Method and system for quickly deducing damaged state of ship power system

Abstract

The invention belongs to the technical field of ship electric power, and discloses a method and a system for quickly deducing a damaged state of a ship electric power system, and the method comprises the steps: carrying out the modeling of a tree-shaped Bayesian network; and quickly calculating the joint probability of the outage combination. According to the method, a rapid inference algorithm of linear complexity is provided by combining a tree operation structure of a power system, and the exponential complexity O (2n) of a traditional method is reduced to the linear polynomial complexity O (n). The method is realized by calculating the joint probability of the outage combination in a recursion mode: matrix multiplication operation is only required to be performed on each node once, and the total operation times and the node number n are in a linear relation. For an 86-node ship power system, the deduced time consumption is reduced from several hours of a traditional method to less than 0.75 second, and the efficiency is improved by more than 10000 times.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of marine electrical technology, and particularly relates to a method and system for rapidly inferring the damaged state of a marine electrical system. Background Technology

[0002] Shipboard electrical systems are a core component of modern ships, providing power for propulsion systems, living quarters, and other facilities. With the application of new technologies such as all-electric propulsion and integrated power systems, the scale and complexity of shipboard electrical systems continue to increase, encompassing numerous bus nodes, generators, transformers, transmission lines, and load equipment. In the event of unforeseen incidents (such as severe sea conditions, equipment failure, or cyberattacks), shipboard electrical systems may suffer widespread damage, leading to power outages in critical loads (such as navigation systems, communication equipment, and medical facilities), severely threatening ship safety and operational capabilities.

[0003] Rapidly and accurately assessing the extent of damage to a power system after an incident, and identifying which nodes and lines are out of service, is crucial for developing appropriate restoration and repair plans, optimizing resource allocation, and ensuring the rapid restoration of power to critical loads. However, in emergencies, communication networks often suffer severe damage, leading to a significant lack of system status information. The operational status of many nodes cannot be obtained through conventional means such as SCADA systems and phasor measurement units (PMUs), and can only be inferred from a limited number of user alarm messages in areas where communication is unaffected. Under these conditions of extremely sparse and incomplete information, how to quickly and accurately infer the damage status of the entire network, especially critical loads, is a key technical challenge that urgently needs to be addressed in the field of power system emergency management.

[0004] Bayesian networks, as a probabilistic graphical model for handling uncertain information, can express the dependencies between variables and infer the state of unknown variables using limited observation information. They have been applied in fields such as power system fault diagnosis and risk assessment. However, traditional Bayesian network inference algorithms face a severe curse of dimensionality when dealing with large-scale networks, with computational complexity increasing exponentially with the number of nodes, making it difficult to meet the real-time requirements for rapid inference of the damage range of ship power systems.

[0005] (1) Machine learning-based state estimation methods:

[0006] In recent years, machine learning methods such as deep neural networks and support vector machines have been applied to power system state estimation. These methods train models using a large amount of historical data to learn the mapping relationship between measurement data and system state. However, machine learning methods require massive amounts of labeled training data, resulting in high training costs. Furthermore, the models lack physical interpretability, have limited generalization ability under new operating conditions not covered by the training set (such as rare damage patterns), and their reliability is difficult to guarantee. In addition, for ship power systems, the operating modes vary significantly between different ship types and different mission phases, leading to poor model portability.

[0007] (2) Traditional Bayesian network inference methods:

[0008] Bayesian networks represent the conditional dependencies between random variables using directed acyclic graphs, enabling them to infer the probability distributions of other variables based on partial observations (evidence). Theoretically, they are suitable for processing uncertain information. Traditional Bayesian network inference algorithms mainly include:

[0009] Precise inference algorithms, such as variable elimination and junction tree algorithms, calculate posterior probabilities by marginalizing non-query variables. The computational complexity of these algorithms is related to the tree width or the size of the maximum clique in the network; for general Bayesian networks, the time complexity is exponential. For shipboard electrical systems containing dozens or even hundreds of nodes, the computational load is enormous, making real-time inference impossible.

[0010] Approximate inference algorithms, such as Markov Chain Monte Carlo (MCMC) and Variational Inference, approximate the posterior distribution through sampling or optimization methods. While these algorithms can handle large-scale networks, they have slow convergence speeds, and the accuracy and stability of the computational results depend on the number of samples or iterations, and they lack theoretical guarantees.

[0011] Message passing algorithms, such as the Belief Propagation algorithm, iteratively update probabilities by passing messages between nodes. This algorithm is accurate on tree networks, approximate on ring networks, and may not converge or converge to a local optimum for complex topologies.

[0012] Overall, existing Bayesian inference methods fail to fully utilize the physical structural characteristics of power systems (such as tree topologies operating in open loops), and have significant shortcomings in computational efficiency and scalability. Summary of the Invention

[0013] To address the problems existing in the prior art, this invention provides a method for rapidly inferring the damaged state of a ship's electrical system.

[0014] The present invention is implemented as follows. A method for quickly inferring the damaged state of a ship power system includes:

[0015] S1: Tree-shaped Bayesian network modeling;

[0016] S2: Fast calculation of the joint probability of outage combinations.

[0017] Furthermore, the S1 specifically includes:

[0018] (1) According to the open-loop operation structure of the ship power system, a tree-shaped Bayesian network is established. Each distribution network node is mapped to a random variable , and the operating line is mapped to a directed edge. Except for the root node (the uppermost power supply node), each node has only a unique parent node, which is consistent with the characteristic that each node in the power system directly depends on the upstream power supply node.

[0019] (2) Renumber the nodes to ensure that the serial number of the child node is always greater than that of the parent node, that is, if j is the parent node of i, then j < i. This numbering rule facilitates subsequent recursive calculations.

[0020] (3) Define the set of all nodes in the subtree with node i as the root as Sub(i). Define the conditional probability matrix , where represents the parent node of.

[0021] (4) According to the chain rule of the Bayesian network, the joint probability of all random variables is:

[0022] ;

[0023] : The state random variable of the i-th node, taking values 0 (normal) or 1 (outage);

[0024] : the set of parent nodes of;

[0025] N: The total number of network nodes;

[0026] Sub(i): The set of all nodes in the subtree ;

[0027] Furthermore, the S2 specifically includes:

[0028] (1) First, according to the given outage combination , determine the indicator vector of each node;

[0029] (2) Next, the local probability vector Initialize to ;

[0030] (3) Then traverse in reverse order of nodes, for each node Set its parent node Local probability vector Updated to: ;

[0031] The traversal stops when the current node has no parent node;

[0032] (4) Finally, the global root node is obtained. (Right now Local probability vector It is easy to know Based on the law of total probability, a fast inference formula can be derived:

[0033] ;

[0034] in, It is the prior probability of the root node's value;

[0035] During the traversal in step (3), each parent node By all its child nodes The updated process clearly shows that the above procedure implies the following formula:

[0036] ;

[0037] The symbols appearing in the above formula Both the multiplication sign and the multiplication sign refer to the element-wise multiplication of quantities; this formula is the theoretical basis of the recursive process, revealing the relationship between the local probability vectors of the parent node and the child node.

[0038] New parameter definition:

[0039] The k-th outage combination represents a specific outage state configuration of the node set S;

[0040] :node In the shutdown combination The indicator vector below;

[0041] : with nodes The root subtree in the shutdown combination The following state configuration;

[0042] :node The parent node;

[0043] : Out-degree of node A (number of child nodes);

[0044] Furthermore, the proof process for step (4) is as follows:

[0045] Suppose the root node of a subtree is Its out-degree is ,node The child nodes are , with nodes The subtree rooted at node A is denoted as . Assuming a subtree CCP There are nodes, among which Let there be variables to be determined, denoted as . ; Let there be three non-variables to be determined, denoted as... The last one is the root node of the subtree. .

[0046] Without loss of generality, let us assume It is the variable to be determined, and The variable is not to be determined.

[0047] node Let any set of values ​​be set as Variables to be determined and The value of the joint probability expression to be determined is assumed to be... and Specifically, if A is also a variable to be determined, its value in the joint probability expression to be determined is assumed to be... .

[0048] Another objective of this invention is to provide a Bayesian network-based system for rapidly inferring the damaged state of a ship's electrical system, based on the aforementioned method for rapid inference of the damaged state of a ship's electrical system. This system specifically includes:

[0049] The tree-based Bayesian network modeling module abstracts the ship's power system into a tree-based Bayesian network G={N,E,Θ}, where N is the set of nodes representing the bus, E is the set of edges representing the operating route, and Θ is the set of parameters representing the conditional probability matrix. Each node corresponds to a binary random variable, where 0 represents normal operation and 1 represents shutdown. Each edge represents the dependency relationship between parent and child nodes.

[0050] The parameter learning module uses historical running data to determine the conditional probability matrix of the Bayesian network through maximum likelihood estimation; for each node i and its parent node Pa(i), the conditional probability P(Xi|Pa(Xi)) is calculated. This process is completed offline, providing a probabilistic basis for subsequent inference.

[0051] The evidence processing module receives alarm information from the user side and organizes it into an evidence set E={(e1,s1), (e2,s2), ..., (em,sm)}, where ei is the evidence node and si is its shutdown status, 0 or 1; and calculates the joint probability P(E) of the shutdown combination corresponding to the evidence set.

[0052] The fast inference calculation module, for a given set of key nodes C, iteratively calculates the posterior probability P(C combination|E) of each shutdown combination, and uses a recursive method to calculate the joint probability, achieving fast inference with linear complexity;

[0053] The results output module selects the combination of critical nodes that have the highest posterior probability as the most likely range of damage and outputs the inference results for use in formulating recovery plans.

[0054] The ring structure processing module, for power systems with ring structures, adopts a heuristic method based on graph theory and rules to decompose the network topology, extract the tree structure, and infer the node state, thereby realizing the inference of the damage range.

[0055] Another object of the present invention is to provide a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the method for rapidly inferring the damaged state of the ship's electrical system.

[0056] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method for rapidly inferring the damaged state of the ship's electrical system.

[0057] Another objective of this invention is to provide an information data processing terminal, which is used to realize the rapid inference system for the damaged state of the ship's power system based on Bayesian networks.

[0058] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0059] 1. The computational complexity is significantly reduced.

[0060] By incorporating the tree-like operational structure of power systems, a fast inference algorithm with linear complexity is proposed, reducing the exponential complexity O(2^n) of traditional methods to a linear polynomial complexity O(n). This is achieved by calculating the joint probability of outage combinations recursively: only one matrix multiplication operation is required for each node, and the total number of operations is linearly related to the number of nodes n. For an 86-node ship power system, the inference time is reduced from several hours using traditional methods to less than 0.75 seconds, representing an efficiency improvement of over 10,000 times.

[0061] 2. It has good adaptability to the sparsity of alarm information.

[0062] The method of this invention can achieve effective inference even when alarm information is extremely sparse. Tests show that even when evidence nodes account for only about 10% of the total number of nodes, the outage status of critical nodes can still be accurately inferred. This is because the method fully utilizes the probabilistic reasoning ability of Bayesian networks and prior knowledge of the power system (such as the physical law that if the upstream goes out, the downstream will also go out), enabling it to infer the most likely system state from limited information.

[0063] 3. Wide range of applications.

[0064] The method achieves fast computation in univariate polynomial time for any selected combination of key nodes, and has a wider range of applications compared to other fast inference algorithms such as message passing and connection tree algorithms. Furthermore, for power systems with ring structures, a heuristic inference method based on graph theory and rules is proposed, capable of handling multi-path power supply scenarios with an inference accuracy exceeding 90%.

[0065] 4. Supports real-time rolling updates.

[0066] Because of its high computation speed (on the order of seconds), when new alarm information appears, the posterior probability of each outage combination can be quickly recalculated, enabling rolling updates of the damaged area and providing timely basis for recovery and repair plans after damage. This has significant practical value for emergency response scenarios.

[0067] 5. Solid theoretical foundation.

[0068] This invention, based on rigorous probability theory and Bayesian network theory, provides a complete mathematical proof of the recursive formula, including lemma proofs and inductive reasoning. Experimental results perfectly match the theoretical analysis, verifying the correctness of the method. Compared to empirical or heuristic methods, this invention has a solid mathematical foundation and theoretical guarantee.

[0069] 6. Low implementation cost.

[0070] This invention is a pure software algorithm, requiring no additional hardware investment and can be directly integrated into existing ship power monitoring systems. Parameter learning can be completed offline, requiring only a one-time training using historical operating data. The system requires only minimal computing resources during operation, making it particularly suitable for applications with limited computing resources, such as those on ships.

[0071] Currently, there are no mature technologies or products specifically designed for rapid inference of the damaged state of ship electrical systems, either domestically or internationally. This invention is the first to combine tree-based Bayesian networks with the physical characteristics of power systems, achieving rapid and accurate inference under conditions of sparse information, thus filling a technological gap in this field.

[0072] This invention solves the long-standing problem of rapidly and accurately inferring the extent of damage to a large-scale power system under conditions of communication interruption and extreme information loss, providing an effective technical means for power system emergency management.

[0073] Traditional views hold that Bayesian networks are unsuitable for real-time inference in large-scale systems. This invention overcomes this technical bias by fully utilizing the tree-like structure of power systems, achieving fast inference with linear complexity. Attached Figure Description

[0074] Figure 1 This is a flowchart of a method for rapidly inferring the damaged state of a ship's electrical system provided in an embodiment of the present invention.

[0075] Figure 2 This is a flowchart of the fast calculation of joint probability provided in an embodiment of the present invention.

[0076] Figure 3 This is an example diagram illustrating the damage inference results of a power system with a ring structure provided in an embodiment of the present invention.

[0077] Figure 4 This is a comparison chart of the combined probability calculation results of the shutdown combination provided in the embodiments of the present invention.

[0078] Figure 5 This is a comparison chart of the time taken to infer the damaged area using two methods provided in the embodiments of the present invention.

[0079] Figure 6 This is a diagram showing the inference results of the outage range of key nodes provided in the embodiments of the present invention. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0081] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for rapidly inferring the damaged state of a ship's electrical system, the method comprising:

[0082] S1: Tree-based Bayesian network modeling;

[0083] S2: Fast calculation of the combined probability of shutdown combinations.

[0084] S1 specifically includes:

[0085] (1) Based on the open-loop operation structure of the ship's power system, a tree-based Bayesian network is established. Each distribution network node is mapped to a random variable. , the running lines are mapped as directed edges. Except for the root node (the most upstream power supply node), each node has only a unique parent node, which is consistent with the characteristic that each node in the power system directly depends on the upstream power supply node.

[0086] (2) Renumber the nodes to ensure that the serial number of the child node is always greater than that of the parent node, that is, if j is the parent node of i, then j < i. This numbering rule facilitates subsequent recursive calculations.

[0087] (3) Define the set of all nodes in the subtree with node i as the root as Sub(i). Define the conditional probability matrix , where represents the parent node of.

[0088] (4) According to the chain rule of the Bayesian network, the joint probability of all random variables is: ;

[0089] As Figure 2 shown, the specific steps of S2 include:

[0090] (1) First, according to the given outage combination , determine the indicator vector of each node;

[0091] (2) Secondly, initialize the local probability vector to ;

[0092] (3) Then traverse in reverse order of the nodes. For each node , update the local probability vector of its parent node to:

[0093] ;

[0094] Stop traversing until the current node has no parent node;

[0095] (4) Finally, obtain the local probability vector of the global root node (i.e., ). It is easy to know that According to the total probability formula, the fast inference formula can be obtained:

[0096] ;

[0097] Among them, is the prior probability of the root node value;

[0098] During the traversal in step (3), each parent node By all its child nodes The updated process clearly shows that the above procedure implies the following formula:

[0099]

[0100] The symbols appearing in the above formula Both the multiplication sign and the multiplication sign refer to the element-wise multiplication of quantities; this formula is the theoretical basis of the recursive process, revealing the relationship between the local probability vectors of the parent node and the child node.

[0101] Furthermore, the proof process for step (4) is as follows:

[0102] Suppose the root node of a subtree is Its out-degree is ,node The child nodes are , with nodes The subtree rooted at node A is denoted as . Assuming a subtree CCP There are nodes, among which Let there be variables to be determined, denoted as . ; Let there be three non-variables to be determined, denoted as... The last one is the root node of the subtree. .

[0103] Without loss of generality, let us assume It is the variable to be determined, and The variable is not to be determined.

[0104] node Let any set of values ​​be set as Variables to be determined and The value of the joint probability expression to be determined is assumed to be... and Specifically, if A is also a variable to be determined, its value in the joint probability expression to be determined is assumed to be... .

[0105] The fast derivation formula is proved below. To simplify the proof, we will... and All are abbreviated as and Proof:

[0106] ;

[0107] (1) Calculate the terms on the right side of the equation. :

[0108] like It is itself a variable to be determined ( ).at this time Then the conditional probability vector Only with The corresponding component is not 0, this item is Therefore:

[0109] ;

[0110] like It is not itself a variable to be determined ( ),but At this point, we have:

[0111] ;

[0112] In conclusion, .

[0113] (2) Calculate the extended sub-terms on the right side of the equation. :

[0114] like It is itself a variable to be determined ( Therefore and Each matrix has only one non-zero component. When multiplied, only the corresponding column is included in the calculation. Therefore, the following formula holds true:

[0115] ;

[0116] ;

[0117] Otherwise, if It is itself a variable to be determined ( Therefore If all components are 1, then the following equation holds:

[0118] ;

[0119] ;

[0120] (3) Calculate the extended sub-terms on the right side of the equation. :

[0121] use The calculation yielded Before proceeding, a lemma is given (the proof of which is noted in note 1).

[0122] Lemma 1:

[0123] ;

[0124] According to Lemma 1, we have:

[0125] ;

[0126] The above formula applies to non-variables. By enumerating and summing, we can obtain:

[0127] ;

[0128] Simplifying the above equation, we get:

[0129] ;

[0130] Therefore:

[0131] ;

[0132] (4) Calculate the terms on the left side of the equation. :

[0133] If A is the variable to be determined, then ,at this time:

[0134] ;

[0135] At this time Only with The corresponding component is 1, therefore:

[0136] ;

[0137] If A is not the variable to be determined, then ,

[0138] ;

[0139] In conclusion, there is always The proof is complete.

[0140] Note 1: Proof of Lemma 1:

[0141] First, calculate the general joint probability:

[0142] ;

[0143] Based on the above formula, the non-undetermined variables are... Summing yields:

[0144] ;

[0145] Specific application areas or related products of this invention:

[0146] This invention provides a Bayesian network-based system for rapidly inferring the damaged state of a ship's electrical system, based on the aforementioned method for rapid inference of ship electrical system damage. The system specifically includes:

[0147] The tree-based Bayesian network modeling module abstracts the ship's power system into a tree-based Bayesian network G={N,E,Θ}, where N is the set of nodes representing the bus, E is the set of edges representing the operating route, and Θ is the set of parameters representing the conditional probability matrix. Each node corresponds to a binary random variable, where 0 represents normal operation and 1 represents shutdown. Each edge represents the dependency relationship between parent and child nodes.

[0148] The parameter learning module uses historical running data to determine the conditional probability matrix of the Bayesian network through maximum likelihood estimation; for each node i and its parent node Pa(i), the conditional probability P(Xi|Pa(Xi)) is calculated. This process is completed offline, providing a probabilistic basis for subsequent inference.

[0149] The evidence processing module receives alarm information from the user side and organizes it into an evidence set E={(e1,s1), (e2,s2), ..., (em,sm)}, where ei is the evidence node and si is its shutdown status, 0 or 1; and calculates the joint probability P(E) of the shutdown combination corresponding to the evidence set.

[0150] The fast inference calculation module, for a given set of key nodes C, iteratively calculates the posterior probability P(C combination|E) of each shutdown combination, and uses a recursive method to calculate the joint probability, achieving fast inference with linear complexity;

[0151] The results output module selects the combination of critical nodes that have the highest posterior probability as the most likely range of damage and outputs the inference results for use in formulating recovery plans.

[0152] The ring structure processing module, for power systems with ring structures, adopts a heuristic method based on graph theory and rules to decompose the network topology, extract the tree structure, and infer the node state, thereby realizing the inference of the damage range.

[0153] This invention provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method for rapidly inferring the damaged state of the ship's electrical system.

[0154] This invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method for rapidly inferring the damaged state of a ship's electrical system.

[0155] This invention provides an information data processing terminal, which is used to implement the Bayesian network-based rapid inference system for the damaged state of a ship's power system.

[0156] Evidence related to the technical effects obtained by the embodiments of the present invention:

[0157] The theoretical basis for the reliability of this invention includes:

[0158] Mathematical rigor: The recursive algorithm has a rigorous mathematical proof;

[0159] Complexity theory: Linear complexity has an explicit upper bound proof;

[0160] Physical consistency: The algorithm fully utilizes the physical laws of the power system;

[0161] Statistical significance: The test data passed rigorous statistical tests;

[0162] Verifiability: The algorithm's correctness is verified through 200 random tests;

[0163] Scalability: Linear complexity ensures the feasibility of large-scale applications;

[0164] These theoretical bases collectively demonstrate that the performance advantages exhibited by the present invention in the embodiments are reliable and sustainable.

[0165] Repeatable and generalizable.

[0166] Example 1: IEEE 123-node standard case test:

[0167] 1. Test configuration:

[0168] Number of nodes: 123;

[0169] Key nodes: 11 (nodes 3, 16, 21, 39, 51, 56, 60, 66, 73, 82, 116);

[0170] Evidence nodes: 12 (normal operation: 4, 12, 24, 36, 44, 59; out of service: 19, 32, 69, 92, 107, 123);

[0171] Bayesian network parameters: learned in advance using historical data.

[0172] 2. Inference Result:

[0173] Key nodes for normal operation: 3, 16, 21, 39, 51, 56, 60;

[0174] Key points for service disruption: 66, 73, 82, 116;

[0175] Inference time: 0.749 seconds;

[0176] Result Verification: The inference results satisfy the fundamental physical law that if the parent node is out of service, the child node must be out of service; if the child node is operating normally, the parent node must be operating normally; and the direction of probability propagation is consistent with the direction of power flow. The outage scenario diagram is shown below. Figure 6 As shown.

[0177] 3. Performance Comparison:

[0178] Traditional Bayesian inference methods require enumerating 2^(123-12)≈10^33 combinations, resulting in unacceptable computational time costs and making inference impossible. Figure 4 The significant differences in computation time among the three methods are visually demonstrated. The method of this invention avoids the combinatorial explosion problem through recursive computation.

[0179] The method of this invention completes the inference in 0.749 seconds, meeting the real-time requirements.

[0180] Example 2: Verification of algorithm complexity for subnets of different sizes:

[0181] To verify the linear complexity of the algorithm, tests were conducted on IEEE 123 subnets of different sizes:

[0182]

[0183] Test Result Analysis:

[0184] The time taken by the method of this invention remains stable (0.3-0.4 seconds) as the number of nodes increases, verifying the linear complexity characteristic;

[0185] Traditional methods take an exponentially longer time with the number of nodes, exceeding one hour with 35 nodes.

[0186] Performance improvement: For a 30-node system, this invention is about 10,000 times faster than traditional methods. Figure 5 The graphs clearly show that the computation time of traditional methods increases exponentially with the number of nodes, while the method of this invention maintains an approximately linear growth. In an 86-node system, the method of this invention can still complete the inference within 0.75 seconds, meeting the real-time requirements of emergency response for ship electrical systems (typically requiring diagnosis within 3 seconds).

[0187] Example 3: Actual Case Study of Ship Power System:

[0188] 1. Test system configuration:

[0189] A certain type of ship's electrical system:

[0190] Busbar nodes: 84;

[0191] Generators: 8 units;

[0192] Energy storage devices: 12 units;

[0193] Transmission lines: 98;

[0194] Transformers: 20 units;

[0195] Circuit breakers: 122;

[0196] Voltage ratings: 10kV for medium voltage and 220V for low voltage.

[0197] 2. Scenario 1: Safety pump load is damaged:

[0198] Damaged equipment: Safety pump load - D4QF1;

[0199] Inference results: D4QF1 is in a de-energized state, and circuit breaker D4QF1-1 is in an open state;

[0200] Of the 122 circuit breakers in the entire network, 111 were correctly identified.

[0201] Topological inference accuracy: 90.98%.

[0202] 3. Scenario 2: Damage to normal medium-voltage load:

[0203] Damaged equipment: Shipboard electrical distribution and jumper plate #8;

[0204] Inference results: The ship's side power distribution and bridging plate #8 are in a de-energized state, the XK8-5 circuit breaker is in an open state, and the ship's side power distribution and bridging plate #6 are also in a de-energized state (downstream node).

[0205] Of the 122 circuit breakers in the entire network, 111 were correctly identified.

[0206] Topological inference accuracy: 90.98%.

[0207] 4. Comprehensive performance evaluation:

[0208] Inference accuracy: >90%;

[0209] Inference speed: completed in seconds;

[0210] System robustness: It has good adaptability to different damage scenarios;

[0211] Real-time update capability: Supports rolling updates of new alarm information;

[0212] Example 4: Verification of the correctness of joint probability calculation:

[0213] To verify the correctness of the fast algorithm, 200 random tests were conducted on a 30-node subnet:

[0214] Key node ratio: 40% (each node has a 40% probability of becoming a key node);

[0215] Number of tests: 200;

[0216] Comparison method: Traditional enumeration summation method;

[0217] Verification results: In 200 tests, the joint probability results calculated by the method of this invention and the traditional method were completely consistent (with an error of 0), proving the correctness of the algorithm.

[0218] Example 5: Tree-based Bayesian network modeling and fast inference:

[0219] Taking a typical shipboard medium-voltage power supply system as an example, it includes a main generator, busbar sectionalizing switches, distribution cabinets, and terminal loads. Each busbar node is defined as a random variable with a value of 0 (normal) or 1 (out of service), and each distribution line is defined as a directed edge. The dependencies between nodes are determined by collecting long-term operational data, and the conditional probability matrix is ​​obtained using maximum likelihood estimation. During inference, an outage signal for a certain area is input. The system traverses in reverse order starting from the terminal nodes, updating the local probability vector of the parent nodes layer by layer. In one traversal, the combined outage probability of the entire network is obtained, and finally, the most likely area of ​​damage is output.

[0220] In this embodiment, the algorithm was tested on a tree network with 68 nodes. The joint probability calculation time was less than 0.1 seconds, verifying that the recursive calculation has linear complexity. Compared with the traditional full probability enumeration method, the calculation speed of this embodiment is improved by about 20 times, which can meet the real-time requirements of ship electrical system accident diagnosis. The results are attached. Figure 3 As shown in the diagram, the numbers within the nodes represent unique identifiers.

[0221] Example 6: Conditional probability parameter learning process:

[0222] Using one year's worth of historical data recorded by the ship's operation monitoring system, the operational and shutdown status of each bus node is statistically analyzed. For node i and its parent node Pa(i), the number of times node i is out of service is counted under different parent node states, and the conditional probability value is calculated using the maximum likelihood method. To avoid the zero-frequency problem, a smoothing factor of 1 is introduced to ensure the stability of the conditional probability matrix. The resulting parameter set is stored offline to provide probabilistic support for online inference.

[0223] Experiments show that the posterior probability results of the model after parameter learning match the actual maintenance records by more than 90% under the same evidence input, verifying that probability learning based on historical data can truly reflect the system's operating characteristics.

[0224] Example 7: Network decomposition and fusion inference with ring structure:

[0225] For a shipboard power distribution network with a main power supply loop and a backup loop, the set of nodes forming the loop is first identified. Based on the principles of minimum cut set and priority for critical loads, one branch is cut, decomposing the loop structure into two directed spanning trees. A recursive inference process is performed on each subtree to obtain its respective likelihood value, and the posterior distribution of the entire network is obtained through product normalization. Finally, the node combination with the highest posterior probability is selected as the affected area.

[0226] Test results show that when the ring network has 80 nodes, the posterior probability error after fusion is less than 2%, and the inference time is kept within 0.2 seconds, achieving rapid and accurate diagnosis under ring network conditions.

[0227] Example 8: Modular Implementation of the System

[0228] The complete system is built on an industrial computer platform and includes six functional modules: modeling, parameter learning, evidence processing, rapid inference, result output, and ring structure processing. Users input alarm node information through a graphical interface, and the system automatically matches the corresponding probability model, performs recursive inference, and outputs the most likely damaged node and its probability value. All calculations are completed within 0.5 seconds.

[0229] The system operates stably on both simulated ship electrical networks and actual ship experimental platforms, with an inference consistency rate exceeding 95%, effectively supporting rapid decision-making and fault recovery.

[0230] Example 9: Application of computer-readable storage media:

[0231] A computer-readable storage medium is integrated into the ship's monitoring terminal, storing program instructions for a rapid inference method of the ship's electrical system damage status. After the processor invokes the instructions, it automatically completes data reading, network loading, recursive inference, and result output. The system can restart after a power outage to restore the original model parameters and inference records, ensuring data continuity.

[0232] This embodiment demonstrates that the storage medium can effectively support the operation of the algorithm in various embedded hardware, enabling automated damage inference and intelligent decision-making.

[0233] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0234] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for rapidly inferring the damaged state of a ship's electrical system, characterized in that, Based on the open-loop operation structure of the ship's power system, the distribution network nodes are abstracted as random variables with values ​​of zero or one, and the running lines are abstracted as directed edges. Except for the upstream power supply node, each node has only one parent node. Number the nodes such that the number of any child node is greater than the number of its parent node; establish a conditional probability matrix for each node and its set of parent nodes; The chain rule is used to express the joint probability of all random variables as the product of all conditional probabilities. Upon receiving the shutdown combination, a corresponding node indicator vector is constructed, and the local probability vector of each node is initialized to the value distribution without evidence. Starting from the last numbered node, the local probability from its child nodes is traversed in reverse order. For each node, the local probability is reduced according to the conditional probability and multiplied with the corresponding element of the current local probability of the parent node to complete one upward propagation. When traversing to the point where there is no parent node, the local probability vector of the root node is obtained. According to the law of total probability, the prior probability of the root node is multiplied by the local probability vector to obtain the joint probability of the shutdown combination. Then, the posterior probability is calculated for each candidate combination of the given set of key nodes, and the shutdown combination with the largest posterior probability is selected as the most likely range of damage. Thus, the inference is completed quickly with a complexity that is linearly related to the number of nodes.

2. The method according to claim 1, characterized in that, The prior probability of the root node is obtained by statistical analysis of historical running data. The element-wise multiplication of the local probability vector is defined as multiplying the elements with the same index and then replacing the corresponding element value of the parent node.

3. The method according to claim 1, characterized in that, When updating the parent node of the subtree root node multiple times during the upward propagation process, the commutative and associative laws that are independent of the order of the child nodes are used to achieve one traversal aggregation, ensuring that the joint probability calculation of the entire network can be completed in only one bottom-up propagation, so that the computational complexity is linearly proportional to the number of nodes.

4. A parameter learning method for conditional probabilities of Bayesian networks in ship electrical systems, characterized in that, Historical operational data is collected and event timestamps are aligned and missing values ​​are imputed. For each node and its set of parent nodes, the frequency of occurrence of the node taking zero and one in the state of the parent node is enumerated. The frequency estimate of each entry of the conditional probability matrix is ​​obtained by using the maximum likelihood criterion, and the zero-frequency entries are smoothed by Laplace with a fixed smoothing coefficient to obtain the parameter set for online inference.

5. The method according to claim 4, characterized in that, The event timestamps are aligned and aggregated in milliseconds. The node status is determined by majority voting among multiple records within the window, while records outside the window do not participate in the conditional probability statistics of that window.

6. A method for rapid inference of the damage state of a ship's electrical system containing a ring structure, characterized in that, Based on graph theory and rules, the loop-containing topology is decomposed. Cut sets that satisfy the minimum cutting scale and maintain the power supply priority of important loads are selected to disconnect the loop and extract several directed spanning trees. For each directed spanning tree, the fast inference described in claim 1 is performed independently to obtain the evidence likelihood corresponding to each tree. The evidence likelihoods of each tree are then multiplied and the product is normalized with a normalization constant to obtain the posterior probability distribution of the entire network. The outage combination with the largest posterior probability is output as the damage range of the loop-containing network.

7. The method according to claim 6, characterized in that, The selection of the cut set takes the number of cut branch entries as the first optimization objective and the minimization of the risk of voltage exceedance of the critical bus as the second optimization objective. The second optimization objective is represented by a weighted summation of the exceedance probability obtained from historical power flow statistics.

8. A rapid inference system for damaged states of a ship's electrical system based on Bayesian networks, characterized in that, It includes a tree-based Bayesian network modeling module, a parameter learning module, an evidence processing module, a fast inference calculation module, a result output module, and a ring structure processing module; The tree-based Bayesian network modeling module is used to construct a network by using the bus as a set of nodes, the running lines as a set of edges, and the conditional probability matrix as a set of parameters. The parameter learning module is used to obtain the conditional probability matrix by maximum likelihood with Laplace smoothing; the evidence processing module is used to organize the alarm information from the user side into an evidence set of the node and its shutdown value and calculate the joint probability of the shutdown combination corresponding to the evidence. The fast inference calculation module is used to perform bottom-up recursive aggregation on the candidate combinations of the key node set as described in claim 1 and output the posterior probability of each combination; The results output module is used to select the shutdown combination with the highest posterior probability as the most likely range of damage and provide a list of nodes for recovery; The ring structure processing module is used to perform topological decomposition, tree structure extraction and probability fusion according to the method described in claim 6 when a ring is detected.

9. The system according to claim 8, characterized in that, The fast inference calculation module aggregates the local probabilities from each child node by multiplying corresponding elements in a bottom-up traversal. After the traversal, it performs an inner product of the root node's prior probability and the root node's local probability vector to obtain the joint probability. Then, it normalizes the prior of the key node combination with the joint probability to obtain the posterior probability ranking result.

10. A computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the rapid inference method for the damaged state of a ship's electrical system as described in claim 1.

Images