Ship start-stop control method based on working condition prediction

Abstract

The invention discloses a ship start-stop control method based on working condition prediction, and the method comprises the steps: building a dynamic model of working condition prediction according to ship operation data, and obtaining a predicted working condition of a ship according to the model; and setting a target function and a constraint condition of power distribution according to the predicted working condition, finding an optimal power distribution path in the complex network in combination with a quantum annealing algorithm to obtain an optimized power distribution result, and controlling the start and stop operation of the ship according to the power distribution result. The invention provides a ship start-stop control method based on working condition prediction, which comprises the following steps: predicting future working conditions of a ship through a dynamic model, and constructing a target function and constraints to optimize global energy consumption and network topology; an optimal power path is searched by adopting a quantum annealing algorithm, so that equipment power is accurately distributed, ship start and stop are adjusted in real time, energy consumption and abrasion are reduced, the overall efficiency of a power system is improved, and the problem that the operation efficiency of ship start and stop is difficult to improve can be solved.

Description

Technical Field

[0001] The present invention relates to the field of automation technology, and particularly to a ship start-stop control method based on working condition prediction. Background Art

[0002] Controlling the start and stop operations of a ship is an important part of the automation system. It can realize the automatic start and stop of the ship's power system through instructions and algorithms, improving the automation level of the ship; a reasonable start-stop control strategy can optimize the performance of the ship's power system, reduce unnecessary energy consumption, and improve the operation efficiency of the ship. For example, when the ship is docking or waiting for a long time, by timely shutting down the power system, fuel consumption and emissions can be reduced. The current ship start-stop operations mainly rely on manual judgment, simple sensor feedback, and rule-based control systems; the crew needs to comprehensively consider the ship's operating status, navigation requirements, and environmental conditions, and manually control the start and stop of the power system; the traditional system only uses simple sensors to monitor key parameters such as engine temperature and oil pressure, and once the parameters reach the preset threshold, the start and stop are automatically triggered; some ships use rule-based control systems to judge the start and stop timing according to preset logic and safety requirements to achieve intelligent management of the power system.

[0003] However, manual judgment of ship start and stop is time-consuming and relies on experience, making it difficult to respond quickly to the rapidly changing operating status; simple sensors only monitor limited parameters and are difficult to comprehensively understand the complex situation of the power system; at the same time, the rule-based control system lacks real-time learning and adaptive capabilities and cannot flexibly adjust the strategy according to real-time data; therefore, the current ship start-stop control means are unable to improve the operation efficiency and urgently need to be innovated to achieve more intelligent and efficient start-stop management to ensure the safety and economy of ship operation. Summary of the Invention

[0004] The present invention provides a ship start-stop control method based on working condition prediction to solve the problem of difficult to improve the operation efficiency of ship start and stop.

[0005] To achieve the above object, the present application provides a ship start-stop control method based on working condition prediction, including: Obtain the current ship operation data from the ship's power system; According to the current ship operation data, perform working condition prediction through a dynamic model to obtain the predicted working condition of the ship; wherein, the dynamic model is established by encoding the network state of the power system according to historical ship operation data and capturing the network dynamic characteristics through quantum computing; Establish an objective function and a constraint set for power distribution according to the predicted working condition; wherein, the objective function is established by minimizing the global energy loss value, maximizing the network topological entropy, and suppressing the overload of the target node; Find the optimal power distribution path in the topological structure of the power system according to the objective function and the constraint set, in combination with the quantum annealing algorithm; Decode the path of the optimal power distribution path and map the device power to obtain the power distribution result, and control the start and stop operations of the ship according to the power distribution result.

[0006] Through the dynamic model, the present invention can accurately predict future working conditions based on the current ship operation data, providing a scientific basis for optimizing start and stop operations; among them, the dynamic model uses historical ship operation data for learning and encodes the network state of the power system. This process can discover the historical laws and characteristics of ship operation, providing a solid foundation for working condition prediction; moreover, quantum computing has powerful parallel processing capabilities and computing speeds, and can efficiently capture and analyze the network dynamic characteristics of the power system. The established objective function aims to minimize global energy loss, maximize network topology entropy, and prevent node overload, ensuring that the power system operates in an efficient and safe state. The quantum annealing algorithm is used to find the optimal power distribution path in the topological structure of the power system. This algorithm is efficient and can handle complex optimization problems to ensure finding the optimal solution. Through path decoding and device power mapping, the optimal power distribution path can be converted into specific device power distribution results. This process ensures that each device obtains the most suitable power for its operating state, avoiding the situation of excessive or insufficient power, thereby improving the efficiency of the entire power system. Finally, the start and stop operations of the ship are adjusted in real time according to the optimal power distribution result, which can avoid unnecessary energy consumption and wear and improve the operating efficiency.

[0007] Compared with the prior art, the present invention predicts the future working conditions of the ship through a dynamic model, constructs an objective function and constraints to optimize the global energy consumption and network topology; uses the quantum annealing algorithm to find the optimal power path, so as to accurately distribute the device power, adjust the start and stop of the ship in real time, reduce energy consumption and wear, and improve the overall efficiency of the power system. Therefore, it can solve the problem of difficult to improve the operating efficiency of ship start and stop.

[0008] As a preferred solution, according to the current ship operation data, perform working condition prediction through a dynamic model to obtain the predicted working conditions of the ship, specifically: Perform data aggregation and weighted calculation on the current ship operation data according to the dynamic model to obtain a time series enhanced embedding and a dynamic weight matrix; Use quantum technology to put the high-betweenness nodes in the ship operation data into a superposition state, and control the high-betweenness nodes to perform quantum jumps and classical transfers according to the time series enhanced embedding and the dynamic weight matrix to obtain a transition probability matrix; Perform path sampling according to the transition probability matrix, and calculate the predicted working conditions of the ship by combining the data obtained from path sampling and the time series enhanced embedding.

[0009] In this preferred solution, the introduction of the dynamic weight matrix enables the model to capture the changes in the ship's operating state over time. Using quantum technology to put high-betweenness nodes into superposition states can make full use of the parallel processing ability of quantum computing and improve the computing efficiency, which is of great significance for processing large-scale and highly complex ship operation data. Moreover, the combination of quantum jumps and classical transfers enables the model to explore more possibilities between the ship's operating states, thus more comprehensively reflecting the dynamic behavior of the system.

[0010] As a preferred solution, data aggregation and weighted calculation are performed on the current ship operation data according to the dynamic model to obtain temporal enhanced embeddings and a dynamic weight matrix, specifically as follows: Aggregate the adjacent node data of each node in the current ship operation data according to the dynamic model to obtain node embedding vectors; According to the node embedding vectors, in combination with a preset learnable parameter matrix, calculate query vectors, key vectors, and value vectors; According to the query vectors, key vectors, and value vectors, calculate the temporal enhanced embeddings of all nodes in the dynamic system through a temporal attention algorithm; Perform attention weighting and normalization processing on the original edge weights to obtain a dynamic weight matrix; wherein, the original edge weights are the dynamic weight values assigned to the connection relationships between mechanical components in the dynamic system.

[0011] In this preferred solution, the node embedding vectors, as representations of node features, can capture the uniqueness of nodes and their position information in the system. This feature expression ability enables the model to more accurately understand the roles and functions of nodes and provide accurate inputs for subsequent calculations. Perform attention weighting and normalization processing on the original edge weights to obtain a dynamic weight matrix. This process not only retains the static information of the connection relationships between mechanical components but also reflects the changes in these relationships over time and operating states through dynamic weight assignment.

[0012] As a preferred solution, path sampling is performed according to the transition probability matrix, and the predicted operating conditions of the ship are calculated by combining the data obtained from path sampling and the temporal enhanced embeddings, specifically as follows: Perform path sampling according to the transition probability matrix to obtain a path distribution and node access frequencies; According to the temporal enhanced embeddings and the path distribution, fuse the path similarity and temporal embedding features through Gaussian process regression to obtain the initial predicted operating conditions of the ship; Perform node marking and path thickening on the initial predicted operating conditions according to the node access frequencies to obtain the predicted operating conditions with a dynamic topology graph.

[0013] This preferred solution performs path sampling through a transition probability matrix, which can capture the transition rules between the operating states of the ship, helping to reflect the dynamics and uncertainties of the system. Gaussian process regression can fuse path similarity and temporal embedding features, which respectively reflect the characteristics of ship operation from the path level and the temporal level, helping to improve the accuracy and robustness of prediction. The dynamic topology map not only provides a comprehensive view of the ship's operating state, but also can mark nodes and thicken paths according to the node access frequency, which helps the system quickly identify key states and paths, and thus make more informed decisions.

[0014] As a preferred solution, the transition probability matrix is specifically:

[0015] where is the probability value, is the quantum unitary operator; represents the classical transition weight from node to node at time ; is the sum of the weights of all outgoing edges of node .

[0016] As a preferred solution, the dynamic model includes an input layer, a wandering layer, and an output layer; wherein, the input layer is established by capturing the temporal enhanced embedding and dynamic weight matrix of ship operation data through temporal graph convolutional encoding and attention mechanism; the wandering layer is established by calculating the mixed probability according to the temporal enhanced embedding and dynamic weight matrix, and performing quantum leap and classical transition through the mixed probability to generate a wandering path sequence with temporal tags; the output layer is established by predicting the future working condition probability by fusing path similarity and embedding features through Gaussian process regression according to the wandering path sequence.

[0017] Through temporal graph convolutional encoding, the input layer of this preferred solution can effectively extract temporal features from historical ship operation data, which are crucial for understanding the dynamic changes of ship operation states. The wandering path sequence with temporal tags generated by the wandering layer not only retains the temporal information of the ship operation state, but also reflects the potential connections between states through the diversity of wandering paths, which is of great significance for understanding the complexity of ship operation and predicting future working conditions. The output layer uses Gaussian process regression to predict the future working condition probability by combining path similarity and embedding features. This method performs well in dealing with small sample data and non-linear relationships, and has high prediction accuracy and robustness.

[0018] As a preferred solution, the objective function is established by minimizing the global energy loss value, maximizing the network topological entropy, and suppressing the overload of target nodes, specifically as follows: Taking minimizing the global energy loss as the objective, a first sub-objective function is established according to the node quantum state overlap degree in the current ship operation data; Taking maximizing the network topological entropy as the objective, a second sub-objective function is established by suppressing the load of high-degree nodes according to the node degree and betweenness centrality parameters; A third sub-objective function is established by imposing a non-linear penalty on the target risk nodes according to the predicted working conditions; The objective function is established according to the first sub-objective function, the second sub-objective function, and the third sub-objective function.

[0019] In this preferred solution, the first sub-objective function is dedicated to minimizing the global energy loss. By accurately calculating the node quantum state overlap degree in the current ship operation data, energy can be allocated and managed more effectively, thereby reducing unnecessary energy waste. The second sub-objective function aims to maximize the network topological entropy, which helps to improve the stability and robustness of the network structure; by reasonably suppressing the load of high-degree nodes, key nodes in the network can be prevented from failing due to overload, thus ensuring the normal operation of the entire network. The third sub-objective function can more effectively identify and manage potential risks by imposing a non-linear penalty on the target risk nodes in the predicted working conditions.

[0020] As a preferred solution, the first sub-objective function is specifically as follows:

[0021] Among them, is the power flow of edge The edge represents the channel of power flow or energy transmission; is the capacity of the edge; is the quantum state overlap degree of node The node represents the starting point or ending point of power flow or energy transmission; E is the set of edges.

[0022] As a preferred solution, the constraint set is composed of hard constraints and soft constraints; Among them, the hard constraints are established according to the load power to limit the power change value and according to the node power to limit the equipment capacity; The soft constraints are established by limiting the time series entropy change value of the power system according to the preset topological entropy change threshold.

[0023] This preferred solution can prevent the system from becoming unstable due to excessive power fluctuations by restricting the range of power changes; ensuring that the power output of each device does not exceed its design capacity can prevent device overload, extend the service life of the device, and improve the overall reliability of the system. Therefore, the setting of hard constraints helps prevent the system from entering a dangerous state. By restricting the range of change of the temporal entropy, it can ensure that the system maintains a certain stability and predictability during the optimization process. Soft constraints are more flexible than hard constraints and can, to a certain extent, allow small changes in the system state to adapt to fluctuations in external conditions.

[0024] As a preferred solution, perform path decoding and device power mapping on the optimal power distribution path to obtain the power distribution result, specifically: Perform binary decoding and power mapping on the optimal power distribution path to obtain the initial power distribution result; According to a preset method, adjust the power distribution of conflicting nodes in the initial power distribution result through topological dependency verification, and perform proportional scaling or load reduction on the initial power distribution result according to the current total load of the dynamic system to obtain the power distribution result.

[0025] The topological dependency verification of this preferred solution ensures the rationality of the power distribution result, avoids power distribution errors caused by node conflicts, and guarantees data accuracy. Performing proportional scaling or load reduction on the initial power distribution result according to the current total load of the dynamic system enables the power distribution result to dynamically adapt to the actual load requirements of the system. This flexibility helps maintain the stability and efficiency of the system when the load changes and improves the adaptability of the system. Description of the Drawings

[0026] Figure 1 is a schematic flowchart of a ship start-stop control method based on working condition prediction provided by an embodiment of the present application; Figure 2 is a schematic structural diagram of a ship start-stop control device based on working condition prediction provided by an embodiment of the present application. Detailed Embodiments

[0027] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0028] In the description of this application, it should be understood that the terms "first", "second", and "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first", "second", and "third" may explicitly or implicitly include one or more of such features. In the description of this application, unless otherwise specified, the meaning of "several" is two or more.

[0029] Embodiment 1: Please refer to Figure 1 , an embodiment of this application provides a ship start-stop control method based on working condition prediction, including S1 to S5, and the specific implementation steps are as follows: S1. Obtain the current ship operation data from the power system of the ship; Specifically, step S1 of the embodiment of this application is as follows: In the power system of the ship, use a data acquisition and monitoring system (SCADA) or other preset methods to collect the current ship operation data in real time, including equipment operation data, time-varying network snapshots, node characteristics, and time series data, etc., and store the current ship operation data in a database.

[0030] Among them, the current ship operation data includes the rotational speed, temperature, and pressure of physical equipment, power transmission efficiency, loss, and the energy and information flow records between equipment, etc., which are used to reflect the equipment status.

[0031] The network snapshot includes the topological structure of the power system that changes over time, including nodes and connections, which is used to reflect the state changes.

[0032] The node characteristics include node type, capacity, and performance parameters, which are used to help the model understand the behavior.

[0033] The time series includes the state data of the system, equipment, and functional modules arranged in chronological order, which is used to adapt the model.

[0034] S2. According to the current ship operation data, perform working condition prediction through a dynamic model to obtain the predicted working condition of the ship; among them, the dynamic model is established by encoding the network state of the power system based on historical ship operation data and capturing the network dynamic characteristics through quantum computing.

[0035] Step S2 of the embodiment of this application includes S2.1 to S2.4, and specifically is as follows: S2.1. Collect historical ship operation data from the power system of the ship, such as the operation status of physical equipment (such as the main engine, gearbox), the working conditions of functional modules (such as power transmission), and the energy flow and information flow between them; Establish a dynamic model including an input layer, a wandering layer, and an output layer according to the historical ship operation data; Among them, the input layer is established by capturing the temporal enhanced embedding and dynamic weight matrix of ship operation data through temporal graph convolutional encoding and attention mechanism. Specifically: in the input layer, a temporal graph convolutional network (TGCN) is used to encode the snapshot sequence of the time-varying complex network, and node embedding vectors are generated by aggregating the neighbor information of each node (considering time dependence). At the same time, a temporal attention mechanism is used to weight the historical embedding sequence to obtain the temporal enhanced embedding and dynamic weight matrix, and these results are directly used as the input of the subsequent walk layer; in addition, the "time-varying complex network" refers to the topological structure that changes dynamically with time in the ship power system; The walk layer is established by calculating the hybrid probability based on the temporal enhanced embedding and dynamic weight matrix, and performing quantum jumps and classical transfers through the hybrid probability to generate a sequence of walk paths with temporal tags. Specifically: in the walk layer, a quantum-classical hybrid walk is implemented by fusing quantum computing and classical random walk strategies, that is, first placing the high-betweenness nodes in a quantum superposition state, and then each step of the walk performs a quantum jump regulated by the dynamic weight matrix with probability p, or a classical transfer regulated by the temporal enhanced embedding with probability 1 - p. By alternately applying these two transfer methods, walk paths are generated and the node access sequence is recorded, providing input for the output layer; The output layer is established by predicting the future working condition probability by fusing path similarity and embedding features through Gaussian process regression based on the sequence of walk paths. Specifically: in the output layer, Gaussian process regression (GPR) technology is used to predict the working condition probability for the next T steps by combining path similarity and node embedding features. Finally, this layer outputs the probability distribution of the working condition and visualization results, such as dynamic topology maps, etc.

[0036] In this embodiment S2.1, through temporal graph convolutional encoding, the input layer can effectively extract temporal features from historical ship operation data, which are crucial for understanding the dynamic changes of the ship operation state. The sequence of walk paths with temporal tags generated by the walk layer not only retains the temporal information of the ship operation state, but also reflects the potential connections between states through the diversity of the walk paths, which is of great significance for understanding the complexity of ship operation and predicting future working conditions. The output layer uses Gaussian process regression to predict the future working condition probability by combining path similarity and embedding features, and this method performs well in dealing with small sample data and non-linear relationships, with high prediction accuracy and robustness.

[0037] S2.2. In the input layer of the dynamic model, based on the current ship operation data, aggregate the adjacent node data of each node to obtain the node embedding vector , such as the main engine node embedding is: ; Based on the node embedding vectors and combined with a preset learnable parameter matrix, the query vector q, the key vector and the value vector are calculated; Based on the query vector q, the key vector and the value vector , the temporal enhanced embeddings of all nodes in the dynamic system are calculated through the temporal attention algorithm ; The original edge weights are subjected to attention weighting and normalization to obtain the dynamic weight matrix ; Among them, the original edge weight is the dynamic weight value assigned to the connection relationship between mechanical components in the dynamic system, such as the transmission efficiency from the main engine to the gearbox.

[0038] Among them, the node embedding vector is the final output of the multi-layer temporal graph convolutional network, which is generated by progressively aggregating neighbor information; in the process of obtaining , there is a single-layer output :

[0039] Among them, the query vector q, the key vector and the value vector are respectively: ; ;

[0040] Among them, is the set of direct neighbor nodes of node v, is the direct neighbor node of node v, is the edge weight matrix of the layer, is the learnable parameter matrix; is the activation function (such as ReLU, and ReLU is short for Rectified Linear Unit, that is, "Rectified Linear Unit", which is an activation function widely used in artificial neural networks); is the node embedding vector of the previous layer; , and are the learnable parameter matrices; is used to represent the interval between the historical time step and the current time step when calculating the key vector and the value vector .

[0041] In the present embodiment S2.2, the node embedding vector, as the representation of node features, can capture the uniqueness of the node and the position information in the system. This feature expression ability enables the model to more accurately understand the role and function of the node, providing accurate input for subsequent calculations. The original edge weights are subjected to attention weighting and normalization to obtain a dynamic weight matrix. This process not only preserves the static information of the connection relationships between mechanical components but also reflects the changes of these relationships over time and operating states through dynamic weight assignment.

[0042] S2.3. In the random walk layer of the dynamic model, high-betweenness nodes (such as the main engine) in the ship operation data are put into the quantum superposition state through quantum technology, and the embedding is enhanced according to the time series and the dynamic weight matrix to control the high-betweenness nodes to perform quantum jumps and classical transfers, obtaining a transition probability matrix ; where, at each step of the random walk, a quantum jump is performed with a probability (regulated by ), and a classical transfer is performed with a probability of 1 - (regulated by ); Path sampling is performed according to the transition probability matrix to obtain a path distribution and the node access frequency .

[0043] Among them, the transition probability matrix is:

[0044] Among them, is a probability value, is a quantum unitary operator; represents the classical transfer weight from node to node at time ; is the sum of the weights of all outgoing edges of node .

[0045] In the present embodiment S2.3, the introduction of the dynamic weight matrix enables the model to capture the changes in the ship operation state over time. Using quantum technology to put high-betweenness nodes into the superposition state can make full use of the parallel processing ability of quantum computing and improve the computing efficiency, which is of great significance for processing large-scale and highly complex ship operation data. Moreover, the combination of quantum jumps and classical transfers enables the model to explore more possibilities between ship operation states, thus more comprehensively reflecting the dynamic behavior of the system; In addition, path sampling through the transition probability matrix can capture the transition laws between the operating states of the ship, which helps to reflect the dynamics and uncertainties of the system.

[0046] S2.4. In the output layer of the dynamic model, according to the temporal enhanced embedding and the path distribution , the path similarity (i.e., path similarity) and the temporal embedding features are fused through Gaussian process regression (GPR) to obtain the initial predicted operating conditions of the ship; According to the node access frequency the initial predicted operating conditions are marked with nodes and the paths are thickened to obtain the predicted operating conditions with a dynamic topology graph.

[0047] Among them, the initial predicted operating conditions are:

[0048] Among them, represents the posterior probability distribution of the predicted operating conditions under the condition of the given path distribution and the node access frequency ; represents the Gaussian process, which is jointly defined by the mean function and the covariance function .

[0049] In the present embodiment S2.4, Gaussian process regression can fuse the path similarity and the temporal embedding features. These two features respectively reflect the characteristics of the ship's operation from the path level and the temporal level, which helps to improve the accuracy and robustness of the prediction. The dynamic topology graph not only provides a comprehensive view of the ship's operating state, but also can mark the nodes and thicken the paths according to the node access frequency, which helps the system to quickly identify the key states and paths, so as to make more informed decisions.

[0050] S3. Establish the objective function and constraint set for power distribution according to the predicted operating conditions; among them, the objective function is established by minimizing the global energy loss value, maximizing the network topology entropy and suppressing the overload of the target node.

[0051] Step S3 of the embodiment of the present application includes S3.1 to S3.2, specifically: S3.1. The objective function is established by minimizing the global energy loss value, maximizing the network topology entropy and suppressing the overload of the target node, specifically: With the goal of minimizing the global energy loss, a first sub-objective function is established according to the node quantum state overlap degree in the current ship operation data. In addition, the classical optimization in the limitations of traditional methods only considers explicit parameters (such as resistance, distance) and ignores implicit coupling relationships, while the first sub-objective function Introduce "soft" topological constraints through the degree of quantum state overlap, enabling the optimizer to actively explore efficient paths that are not explicitly modeled.

[0052] Aiming to maximize the network topological entropy, suppress the load of high-degree nodes according to the node degree and betweenness centrality parameters, and establish a second sub-objective function . In addition, traditional resilience optimization only considers connectivity and ignores the impact of subgraph structure complexity on fault diffusion, while the second sub-objective function combines topological entropy and betweenness centrality, optimizing both global connectivity and local robustness.

[0053] Apply a non-linear penalty to the target risk nodes according to the predicted working conditions, and establish a third sub-objective function . In addition, traditional optimization is based on historical data and cannot handle sudden working conditions such as sudden equipment failures, while the third sub-objective function can avoid the passive response mode of "fault - protection" by reducing the power of high-risk nodes in advance.

[0054] According to the first sub-objective function , the second sub-objective function and the third sub-objective function establish the objective function .

[0055] The first sub-objective function is:

[0056] The second sub-objective function is:

[0057] The third sub-objective function is: ;

[0058] The objective function is

[0059] where is the power flow of edge , and the edge represents the channel of power flow or energy transmission; is the capacity of the edge; is the degree of quantum state overlap of node , and the node represents the starting or ending point of power flow or energy transmission; E is the set of edges; is the set of nodes, is the degree of the node , is the betweenness centrality , is the topological entropy of the subgraph centered at is the risk sensitivity coefficient is the quadratic penalty term is the node at the prediction working condition at the moment is the failure prediction probability , and are dynamic weights, which can be adjusted online through reinforcement learning.

[0060] In Embodiment S3.1, the first sub-objective function is dedicated to minimizing the global energy loss. By accurately calculating the quantum state overlap degree of each node in the current ship operation data, energy can be more effectively allocated and managed, thereby reducing unnecessary energy waste. The second sub-objective function aims to maximize the network topological entropy, which helps to improve the stability and robustness of the network structure; by reasonably suppressing the load of high-degree nodes, key nodes in the network can be prevented from failing due to overload, thus ensuring the normal operation of the entire network. The third sub-objective function can more effectively identify and manage potential risks by imposing non-linear penalties on the target risk nodes in the prediction working condition.

[0061] S3.2. Establish a constraint set according to the hard constraints and soft constraints; Among them, the hard constraints are established based on the prediction working condition, according to the load power to limit the power change value and according to the node power to limit the equipment capacity, including power balance constraints, equipment capacity constraints and topological connectivity constraints; The soft constraints are established based on the prediction working condition, according to the preset topological entropy change threshold to limit the time series entropy change value of the dynamic system, including quantum path probability constraints and time series entropy smoothing constraints.

[0062] Among them, the power balance constraint is:

[0063] The equipment capacity constraint is:

[0064] The topological connectivity constraint is:

[0065] The quantum path probability constraint is:

[0066] The time - series entropy smoothing constraint is:

[0067] Wherein, is the power generation power of node , is the load power, is the set of nodes; and are the upper limit and lower limit of node power respectively, is the power of node ; is the set of critical edges, such as the power transmission path of the main engine gearbox; is the power value transmitted from node to node , such as the mechanical power output from the main engine to the gearbox; is the set of all possible power paths, is the probability weight of path ; I(-) is a binary function, which takes the value of 1 when the condition is met, otherwise 0; is the safe sub - graph under the predicted working condition, is the minimum safe path probability threshold; is the topological entropy of the system topology graph at time step , is the topological entropy of the previous time step , is the total number of time steps, is the topological entropy change threshold.

[0068] In this embodiment, S3.2 can prevent the system from becoming unstable due to excessive power fluctuations by restricting the change range of power; ensuring that the power output of each device does not exceed its design capacity can prevent device overload, extend the service life of the device, and improve the overall reliability of the system. Therefore, the setting of hard constraints helps to prevent the system from entering a dangerous state. By restricting the change range of time - series entropy, it can ensure that the system maintains a certain stability and predictability during the optimization process. Soft constraints are more flexible than hard constraints and can allow small changes in the system state to a certain extent to adapt to fluctuations in external conditions.

[0069] S4. According to the objective function and the constraint set, combined with the quantum annealing algorithm, find the optimal power distribution path in the topological structure of the dynamic system.

[0070] Specifically, step S4 of the embodiment of the present application is as follows: The objective function is converted into the Ising form through quadratic expansion, and the constraint set is converted into quadratic penalty terms; the converted objective function and the constraint set are added together to obtain the total Hamiltonian ; ; According to the total Hamiltonian , the optimal power distribution path is found in the time-varying complex network through the quantum annealing algorithm.

[0071] S5. Decode the path of the optimal power distribution path and map the device power to obtain the power distribution result, and control the start and stop operations of the ship according to the power distribution result.

[0072] Step S5 in the embodiment of this application is specifically as follows: Perform binary decoding and power mapping on the optimal power distribution path to obtain the initial power distribution result. Specifically: the key devices achieve continuous adjustment through 8-bit binary codes, and the auxiliary devices support 16-level discrete power with 4-bit codes. Combine floating-point percentage mapping and device non-linear characteristic correction of data; built-in 2-bit check codes and dynamic proportional distribution mechanism ensure real-time power balance, force adaptation to the minimum starting power and be compatible with the preset hardware interface; finally, dynamically optimize the coding accuracy through load prediction, and combine reinforcement learning to achieve high-precision power distribution adaptable to working conditions to obtain the initial power distribution result; and, the initial power distribution result is specifically presented as a device power distribution table; Adjust the power distribution of the conflict nodes in the initial power distribution result through topology-dependent verification, and scale or reduce the load of the initial power distribution result according to the current total load of the power system to obtain the power distribution result. Specifically: verify whether the power distribution violates physical dependencies (such as power being allocated to the cooling system when the main engine has not started). If a conflict occurs, forcefully correct the power of the core device according to the priority to cover the conflict nodes, and re-calculate the power distribution within the conflict subnet for local re-optimization to obtain the corrected power distribution result; if the current total load of the power system exceeds the limit, scale it proportionally (e.g., multiply all devices by a preset coefficient) or selectively reduce the load (e.g., only reduce the power of non-critical devices); if there is power redundancy, increase the power of expandable devices as needed to obtain the optimized power distribution result.

[0073] Control the start and stop operations of the ship according to the optimized power distribution result.

[0074] The topology dependency check in Embodiment S5 ensures the rationality of the power distribution result, avoids power distribution errors caused by node conflicts, and guarantees data accuracy. Scale or derate the initial power distribution result according to the current total load of the power system, so that the power distribution result can dynamically adapt to the actual load demand of the system. This flexibility helps maintain the stability and efficiency of the system when the load changes, and improves the adaptability of the system.

[0075] Overall, this embodiment has the following beneficial effects: Through the dynamic model, this application can accurately predict future working conditions based on the current ship operation data, providing a scientific basis for optimizing start-stop operations. Among them, the dynamic model uses historical ship operation data for learning and encodes the network state of the power system. This process can discover the historical laws and characteristics of ship operation, providing a solid foundation for working condition prediction. Moreover, quantum computing has powerful parallel processing capabilities and computing speeds, and can efficiently capture and analyze the network dynamic characteristics of the power system. The established objective function aims to minimize the global energy loss, maximize the network topology entropy, and prevent node overload, ensuring that the power system operates in an efficient and safe state. The quantum annealing algorithm is used to find the optimal power distribution path in the power system topology. This algorithm is efficient and can handle complex optimization problems to ensure finding the optimal solution. Through path decoding and device power mapping, the optimal power distribution path can be converted into specific device power distribution results. This process ensures that each device obtains the most suitable power for its operating state, avoiding the situation of excessive or insufficient power, thereby improving the efficiency of the entire power system. Finally, adjust the ship start-stop operation in real time according to the optimal power distribution result, which can avoid unnecessary energy consumption and wear and improve the operation efficiency.

[0076] Embodiment Two: Please refer to Figure 2 , an embodiment of this application provides a ship start-stop control device based on working condition prediction, including a data module 10, a prediction module 20, a calculation module 30, a solution module 40, and a regulation module 50; Among them, the data module 10 is used to obtain the current ship operation data from the power system of the ship; The prediction module 20 is used to perform working condition prediction through a dynamic model according to the current ship operation data to obtain the predicted working conditions of the ship. Among them, the dynamic model is established by encoding the network state of the power system according to historical ship operation data and capturing network dynamic characteristics through quantum computing; The calculation module 30 is used to establish an objective function and a constraint set for power distribution according to the predicted working conditions. Among them, the objective function is established by minimizing the global energy loss value, maximizing the network topology entropy, and suppressing target node overload; A solution module 40, configured to find an optimal power distribution path in the topological structure of the power system according to the objective function and the constraint set, in combination with a quantum annealing algorithm; A regulation module 50, configured to perform path decoding and device power mapping on the optimal power distribution path to obtain a power distribution result, and control the start and stop operations of the ship according to the power distribution result.

[0077] In one embodiment, the data module 10 is specifically: In the power system of a ship, a data acquisition and monitoring system (SCADA) or other preset methods are used to collect current ship operation data in real time, including equipment operation data, time-varying network snapshots, node characteristics, and time series data, etc., and store the current ship operation data in a database.

[0078] Among them, the current ship operation data includes the rotation speed, temperature, and pressure of physical devices, power transmission efficiency and loss, and energy and information flow records between devices, etc., which are used to reflect the device status.

[0079] The network snapshot includes the topological structure of the power system that changes over time, including nodes and connections, which is used to reflect the state changes.

[0080] The node characteristics include node type, capacity, and performance parameters, which are used to help the model understand the behavior.

[0081] The time series includes the state data of the system, equipment, and functional modules arranged in chronological order, which is used to adapt the model.

[0082] In one embodiment, the prediction module 20 includes a model unit, an input unit, a wandering unit, and an output unit; Among them, the model unit is configured to collect historical ship operation data from the power system of the ship, such as the operation status of physical devices (such as the main engine, gearbox), the working conditions of functional modules (such as power transmission), and the energy flow and information flow between them; The model unit is further configured to establish a dynamic model including an input layer, a wandering layer, and an output layer according to the historical ship operation data; Among them, the input layer is established by capturing the time series enhanced embedding and dynamic weight matrix of the ship operation data through time series graph convolution encoding and attention mechanism. Specifically: in the input layer, a time series graph convolutional network (TGCN) is used to encode the snapshot sequence of the time-varying complex network, and the node embedding vector is generated by aggregating the neighbor information of each node (considering time dependence). At the same time, a time series attention mechanism is used to weight the historical embedding sequence to obtain the time series enhanced embedding and dynamic weight matrix, and these results are directly used as the input of the subsequent wandering layer; in addition, the "time-varying complex network" refers to the topological structure in the ship power system that changes dynamically over time; The wandering layer is established by calculating the hybrid probability based on the time-series enhanced embedding and the dynamic weight matrix, and performing quantum jumps and classical transfers through the hybrid probability to generate a sequence of wandering paths with time-series tags. Specifically: in the wandering layer, the quantum computing and classical random walk strategies are fused to implement quantum-classical hybrid walks. That is, the high-betweenness nodes are first placed in a quantum superposition state, and then at each step of the walk, a quantum jump regulated by the dynamic weight matrix is performed with probability p, or a classical transfer regulated by the time-series enhanced embedding is performed with probability 1-p. By alternately applying these two transfer methods, the wandering paths are generated and the node access sequence is recorded, providing input for the output layer; The output layer is established by predicting the probability of future working conditions by fusing path similarity and embedding features through Gaussian process regression based on the sequence of wandering paths. Specifically: in the output layer, the Gaussian process regression (GPR) technique is adopted, and the probability of working conditions in the next T steps is predicted by combining path similarity and node embedding features. Finally, this layer outputs the probability distribution of the working conditions and visualization results, such as dynamic topology maps, etc.

[0083] In this embodiment, the model unit can effectively extract time-series features from historical ship operation data through time-series graph convolution encoding in the input layer. These features are crucial for understanding the dynamic changes in the ship operation state. The sequence of wandering paths with time-series tags generated by the wandering layer not only retains the time-series information of the ship operation state but also reflects the potential connections between states through the diversity of the wandering paths, which is of great significance for understanding the complexity of ship operation and predicting future working conditions. The output layer uses Gaussian process regression to predict the probability of future working conditions by combining path similarity and embedding features. This method performs well in dealing with small sample data and non-linear relationships and has high prediction accuracy and robustness.

[0084] An input unit is used to aggregate the adjacent node data of each node based on the current ship operation data in the input layer of the dynamic model to obtain a node embedding vector , such as the host node embedding being: ; The input unit is also used to calculate the query vector q, the key vector and the value vector based on the node embedding vector and in combination with a preset learnable parameter matrix; The input unit is also used to calculate the time-series enhanced embedding of all nodes in the power system through the time-series attention algorithm based on the query vector q, the key vector and the value vector ; ; The input unit is also used to perform attention weighting and normalization processing on the original edge weights to obtain the dynamic weight matrix ; among them, the original edge weight is the dynamic weight value assigned to the connection relationship between mechanical components in the dynamic system, such as the transmission efficiency from the main engine to the gearbox.

[0085] Among them, the node embedding vector is the final output of the multi-layer temporal graph convolutional network, which is generated by progressively aggregating neighbor information; in the process of obtaining , there is a single-layer output :

[0086] Among them, the query vector q, the key vector and the value vector are respectively: ; ;

[0087] Among them, is the set of direct neighbor nodes of node v, is the direct neighbor node of node v, is the edge weight matrix of the th layer, is the learnable parameter matrix; is the activation function (such as ReLU, and ReLU is short for Rectified Linear Unit, that is, "Rectified Linear Unit", which is an activation function widely used in artificial neural networks); is the node embedding vector of the previous layer; , and are the learnable parameter matrices; is used to represent the calculation of the key vector and the value vector when referring to the interval between the historical time step and the current time step .

[0088] In the input unit of this embodiment, the node embedding vector, as the representation of node features, can capture the uniqueness of the node and its position information in the system. This feature expression ability enables the model to more accurately understand the role and function of the node and provide accurate input for subsequent calculations. The original edge weights are subjected to attention weighting and normalization processing to obtain the dynamic weight matrix. This process not only retains the static information of the connection relationship between mechanical components but also reflects the changes of these relationships over time and operating states through dynamic weight assignment.

[0089] The wandering unit is used in the wandering layer of the dynamic model to put high-betweenness nodes (such as the main engine) in the ship operation data into the quantum superposition state through quantum technology and enhance the embedding according to the time series. And the dynamic weight matrix Control the high-betweenness nodes to perform quantum jumps and classical transfers to obtain the transfer probability matrix ; Among them, each step of wandering executes a quantum jump with probability (regulated by ), and executes a classical transfer with 1 - (regulated by ); The wandering unit is also used to perform path sampling according to the transfer probability matrix to obtain the path distribution and the node access frequency .

[0090] Among them, the transfer probability matrix is:

[0091] Among them, is the probability value, is the quantum unitary operator; represents the classical transfer weight from node to node to node at time ; is the sum of the weights of all outgoing edges of node

[0092] In the wandering unit of this embodiment, the introduction of the dynamic weight matrix enables the model to capture the changes in the ship operation state over time. Using quantum technology to put high-betweenness nodes into the superposition state can make full use of the parallel processing ability of quantum computing and improve the computing efficiency, which is of great significance for processing large-scale and high-complexity ship operation data. Moreover, the combination of quantum jumps and classical transfers enables the model to explore more possibilities between ship operation states, thus more comprehensively reflecting the dynamic behavior of the system; In addition, path sampling through the transfer probability matrix can capture the transfer rules between ship operation states, which helps to reflect the dynamics and uncertainty of the system.

[0093] The output unit is used in the output layer of the dynamic model to fuse the path similarity (i.e., path similarity) and the time series embedding features through Gaussian process regression (GPR) according to the time series enhanced embedding and the path distribution to obtain the initial predicted working condition of the ship; The output unit is also used according to the node access frequency Node marking and path thickening are performed on the initial prediction condition to obtain a prediction condition with a dynamic topology graph.

[0094] Among them, the initial prediction condition is:

[0095] Among them, represents the posterior probability distribution of the prediction condition under the conditions of the given path distribution and the node access frequency ; represents a Gaussian process, which is jointly defined by the mean function and the covariance function ;

[0096] In the output unit of this embodiment, Gaussian process regression can fuse path similarity and temporal embedding features. These two features respectively reflect the characteristics of ship operation from the path level and the temporal level, which helps to improve the accuracy and robustness of prediction. The dynamic topology graph not only provides a comprehensive view of the ship operation state, but also can perform node marking and path thickening according to the node access frequency, which helps the system quickly identify key states and paths, so as to make more informed decisions.

[0097] In one embodiment, the calculation module 30 includes a function unit and a constraint unit; Among them, the function unit is used to establish the objective function by minimizing the global energy loss value, maximizing the network topology entropy and suppressing the overload of the target node. Specifically: Taking minimizing the global energy loss as the goal, a first sub-objective function is established according to the node quantum state overlap degree in the current ship operation data. In addition, classical optimization in traditional methods only considers explicit parameters (such as resistance, distance), ignoring implicit coupling relationships, while the first sub-objective function

[0098] introduces "soft" topological constraints through the quantum state overlap degree, enabling the optimizer to actively explore efficient paths that are not explicitly modeled. Taking maximizing the network topology entropy as the goal, a second sub-objective function is established to suppress the load of high-degree nodes according to the node degree and betweenness centrality parameters.

[0099] In addition, traditional invulnerability optimization only considers connectivity, ignoring the impact of subgraph structure complexity on fault diffusion, while the second sub-objective function In addition, traditional optimization is based on historical data and cannot handle sudden operating conditions such as sudden equipment failures. The third sub-objective function can avoid the passive response mode of "fault - protection" by reducing the power of high-risk nodes in advance.

[0100] According to the first sub-objective function , the second sub-objective function and the third sub-objective function establish the objective function .

[0101] The first sub-objective function is as follows:

[0102] The second sub-objective function is as follows:

[0103] The third sub-objective function is as follows: ;

[0104] The objective function is

[0105] where, is the power flow of edge . The edge represents the channel of power flow or energy transmission; is the capacity of the edge; is the quantum state overlap degree of node . The node represents the starting point or ending point of power flow or energy transmission; E is the set of edges; is the set of nodes, is the degree of node , is the betweenness centrality, is the subgraph topological entropy centered on ; is the risk sensitivity coefficient, is the quadratic penalty term, is the fault prediction probability of node at time in the predicted operating condition, is the risk sensitivity coefficient; , and is a dynamic weight that can be adjusted online through reinforcement learning.

[0106] In the embodiment function unit, the first sub-objective function is dedicated to minimizing the global energy loss. By accurately calculating the quantum state overlap degree of each node in the current ship operation data, energy can be more effectively allocated and managed, thereby reducing unnecessary energy waste. The second sub-objective function aims to maximize the network topology entropy, which helps to improve the stability and robustness of the network structure; by reasonably suppressing the load of high-number nodes, it is possible to avoid the failure of key nodes in the network due to overload, thus ensuring the normal operation of the entire network. The third sub-objective function can more effectively identify and manage potential risks by imposing non-linear penalties on the target risk nodes in the predicted working conditions.

[0107] Among them, the constraint unit is used to establish a constraint set according to hard constraints and soft constraints; Among them, the hard constraints are established based on the predicted working conditions, according to the load power to limit the power change value and according to the node power to limit the equipment capacity, including power balance constraints, equipment capacity constraints, and topological connectivity constraints; The soft constraints are established based on the predicted working conditions, according to the preset topological entropy change threshold to limit the time series entropy change value of the power system, including quantum path probability constraints and time series entropy smoothing constraints.

[0108] Among them, the power balance constraint is:

[0109] The equipment capacity constraint is:

[0110] The topological connectivity constraint is:

[0111] The quantum path probability constraint is:

[0112] The time series entropy smoothing constraint is:

[0113] Among them, is the power generation power of node , is the load power, is the set of nodes; and are the upper limit and lower limit of the node power respectively, is the power of node ; is the set of critical edges, such as the power transmission path of the main engine gearbox; is a node The power value transmitted to the node such as the mechanical power output from the main engine to the gearbox; is the set of all possible power paths, is a path The probability weight of; I(-) is a binary function that takes the value 1 when the condition is satisfied and 0 otherwise; is the safe subgraph under the predicted working conditions, is the minimum safe path probability threshold; is the time step The topological entropy of the system topology graph at this time, is the previous time step The topological entropy of; is the total number of time steps, is the topological entropy change threshold.

[0114] In this embodiment, the constraint unit can prevent the system from becoming unstable due to excessive power fluctuations by restricting the power change range; ensuring that the power output of each device does not exceed its design capacity can prevent device overload, extend the service life of the device, and improve the overall reliability of the system. Therefore, the setting of hard constraints helps to prevent the system from entering a dangerous state. By restricting the change range of the temporal entropy, the stability and predictability of the system during the optimization process can be ensured. Soft constraints are more flexible than hard constraints and can allow small changes in the system state to a certain extent to adapt to fluctuations in external conditions.

[0115] In one embodiment, the solving module 40 is specifically: The objective function is transformed into the Ising form by quadratic expansion and the constraint set is transformed into a quadratic penalty term; the transformed objective function and the constraint set are added together to obtain the total Hamiltonian ; According to the total Hamiltonian an optimal power distribution path is found in the time-varying complex network through the quantum annealing algorithm.

[0116] In one embodiment, the regulation module 50 is specifically: Perform binary decoding and power mapping on the optimal power allocation path to obtain the initial power allocation result. Specifically: The key device realizes continuous adjustment through 8-bit binary codes, and the auxiliary device supports 16-level discrete power with 4-bit codes. Combine floating-point percentage mapping and device to perform non-linear characteristic correction on the data; Built-in 2-bit check codes and dynamic proportional allocation mechanism ensure real-time power balance, forcefully adapt to the minimum startup power and be compatible with the preset hardware interface; Finally, dynamically optimize the coding accuracy through load prediction, and combine reinforcement learning to achieve high-precision power allocation adaptable to working conditions to obtain the initial power allocation result; Moreover, the initial power allocation result is specifically manifested as a device power allocation table; Adjust the power allocation of conflicting nodes in the initial power allocation result through topology-dependent verification, and perform proportional scaling or load reduction on the initial power allocation result according to the current total load of the power system to obtain the power allocation result. Specifically: Verify whether the power allocation violates physical dependencies (such as allocating power to the cooling system when the host is not started). If a conflict occurs, forcibly correct the power of the core device according to the priority to cover the conflicting node, and perform local re-optimization by recalculating the power allocation within the conflicting subnet to obtain the corrected power allocation result; If the current total load of the power system exceeds the limit, perform proportional scaling (such as: multiply all devices by a preset coefficient) or selective load reduction (such as: only reduce the power of non-critical devices); If there is power redundancy, increase the power of expandable devices as needed to obtain the optimized power allocation result.

[0117] Control the start and stop operations of the ship according to the optimized power allocation result.

[0118] The topology-dependent verification in this embodiment S5 ensures the rationality of the power allocation result, avoids power allocation errors caused by node conflicts, and guarantees data accuracy. Performing proportional scaling or load reduction on the initial power allocation result according to the current total load of the power system enables the power allocation result to dynamically adapt to the actual load demand of the system. This flexibility helps to maintain the stability and efficiency of the system when the load changes and improves the adaptability of the system.

[0119] Overall, this embodiment has the following beneficial effects: Through a dynamic model, this application can accurately predict future operating conditions based on current ship operation data, providing a scientific basis for optimizing start-stop operations. Among them, the dynamic model uses historical ship operation data for learning and encodes the network state of the power system. This process can uncover the historical laws and characteristics of ship operation, providing a solid foundation for operating condition prediction. Moreover, quantum computing has powerful parallel processing capabilities and computing speeds, enabling it to efficiently capture and analyze the network dynamic characteristics of the power system. The established objective function aims to minimize global energy loss, maximize network topological entropy, and prevent node overload, ensuring that the power system operates in an efficient and safe state. The quantum annealing algorithm is used to find the optimal power distribution path in the power system topology. This algorithm is efficient and can handle complex optimization problems to ensure finding the optimal solution. Through path decoding and device power mapping, the optimal power distribution path can be converted into specific device power distribution results. This process ensures that each device obtains the power most suitable for its operating state, avoiding situations of excessive or insufficient power, thereby improving the efficiency of the entire power system. Finally, the ship start-stop operations are adjusted in real time according to the optimal power distribution results, which can avoid unnecessary energy consumption and wear and improve the operating efficiency.

[0120] Embodiment 3: The embodiment of this application provides a computer-readable storage medium, which includes a stored computer program. Among them, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the ship start-stop control method based on operating condition prediction described above. Among them, for the ship start-stop control method based on operating condition prediction, if it is implemented in the form of a software functional unit and used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0121] The above are the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A ship start-stop control method based on working condition prediction, characterized in that, Including: Obtaining current ship operation data from the ship's power system; According to the current ship operation data, performing working condition prediction through a dynamic model to obtain the predicted working condition of the ship; wherein, the dynamic model is established by encoding the network state of the power system based on historical ship operation data and capturing network dynamic characteristics through quantum computing; Establishing an objective function and a constraint set for power distribution according to the predicted working condition; wherein, the objective function is established by minimizing the global energy loss value, maximizing the network topology entropy, and suppressing overload of target nodes; According to the objective function and the constraint set, combined with the quantum annealing algorithm, finding the optimal power distribution path in the topological structure of the power system; Performing path decoding and device power mapping on the optimal power distribution path to obtain a power distribution result, and controlling the start and stop operations of the ship according to the power distribution result.

2. The ship start-stop control method based on working condition prediction according to claim 1, characterized in that According to the current ship operation data, performing working condition prediction through a dynamic model to obtain the predicted working condition of the ship, specifically: Performing data aggregation and weighted calculation on the current ship operation data according to the dynamic model to obtain a time series enhanced embedding and a dynamic weight matrix; Using quantum technology to put the high-betweenness nodes in the ship operation data into a superposition state, and controlling the high-betweenness nodes to perform quantum jumps and classical transfers according to the time series enhanced embedding and the dynamic weight matrix to obtain a transition probability matrix; Performing path sampling according to the transition probability matrix, and calculating the predicted working condition of the ship by combining the data obtained from path sampling and the time series enhanced embedding.

3. The method for controlling ship start-stop based on working condition prediction according to claim 2, wherein, Performing data aggregation and weighted calculation on the current ship operation data according to the dynamic model to obtain a time series enhanced embedding and a dynamic weight matrix, specifically: Aggregating the adjacent node data of each node in the current ship operation data according to the dynamic model to obtain a node embedding vector; According to the node embedding vector, combined with a preset learnable parameter matrix, calculating to obtain a query vector, a key vector, and a value vector; According to the query vector, the key vector, and the value vector, calculating the time series enhanced embedding of all nodes in the power system through a time series attention algorithm; Performing attention weighting and normalization processing on the original edge weights to obtain a dynamic weight matrix; wherein, the original edge weights are the dynamic weight values assigned to the connection relationships between mechanical components in the power system.

4. The ship start-stop control method based on working condition prediction according to claim 2, characterized in that, Performing path sampling according to the transition probability matrix, and calculating the predicted working condition of the ship by combining the data obtained from path sampling and the time series enhanced embedding, specifically: Performing path sampling according to the transition probability matrix to obtain a path distribution and a node access frequency; According to the time series enhanced embedding and the path distribution, fusing path similarity and time series embedding features through Gaussian process regression to obtain an initial predicted working condition of the ship; Performing node marking and path thickening on the initial predicted working condition according to the node access frequency to obtain a predicted working condition with a dynamic topology graph.

5. The ship start-stop control method based on working condition prediction according to claim 2, wherein, The transition probability matrix, specifically: ; wherein, is a probability value, is a quantum unitary operator; represents the classical transition weight from node to node at time ; is the sum of the weights of all outgoing edges of node .

6. The ship start-stop control method based on working condition prediction according to claim 1, characterized in that The dynamic model includes an input layer, a wandering layer, and an output layer; Among them, the input layer is established by capturing the temporal enhanced embedding and dynamic weight matrix of ship operation data through temporal graph convolutional encoding and attention mechanism; The walk layer is established by calculating the mixed probability according to the temporal enhanced embedding and dynamic weight matrix, and performing quantum transition and classical transfer through the mixed probability to generate a walk path sequence with temporal tags; The output layer is established by predicting the future working condition probability by fusing path similarity and embedding features through Gaussian process regression according to the walk path sequence.

7. The ship start-stop control method based on working condition prediction according to claim 1, characterized in that, The objective function is established by minimizing the global energy loss value, maximizing the network topological entropy and suppressing the overload of target nodes, specifically: Taking minimizing the global energy loss as the goal, a first sub-objective function is established according to the node quantum state overlap degree in the current ship operation data; Taking maximizing the network topological entropy as the goal, a second sub-objective function is established by suppressing the load of high-degree nodes according to the node degree and betweenness centrality parameters; A third sub-objective function is established by imposing a non-linear penalty on the target risk node according to the predicted working condition; The objective function is established according to the first sub-objective function, the second sub-objective function and the third sub-objective function.

8. The ship start-stop control method based on working condition prediction according to claim 7, characterized in that, The first sub-objective function is specifically: ; Among them, is the power flow of the edge where the edge represents a channel for power flow or energy transmission; is the capacity of the edge; is the degree of quantum state overlap of node where the node represents the starting point or ending point of power flow or energy transmission; E is the set of edges.

9. The ship start-stop control method based on working condition prediction according to claim 1, wherein, The constraint set consists of hard constraints and soft constraints; Among them, the hard constraints are established according to the load power to limit the power change value and according to the node power to limit the equipment capacity; The soft constraints are established by limiting the temporal entropy change value of the power system according to a preset topological entropy change threshold.

10. A ship start-stop control method based on working condition prediction according to claim 1, characterized in that, Performing path decoding and device power mapping on the optimal power distribution path to obtain the power distribution result, specifically: Performing binary decoding and power mapping on the optimal power distribution path to obtain the initial power distribution result; According to a preset method, adjusting the power distribution of conflict nodes in the initial power distribution result through topological dependence verification, and performing proportional scaling or load reduction on the initial power distribution result according to the current total load of the power system to obtain the power distribution result.

Images