Marine ranching device reliability evaluation method and system based on large model

Abstract

The application discloses a kind of based on big model's ocean ranching equipment reliability evaluation method and system, method includes: obtaining the operating parameter of each breeding equipment in ocean ranching, obtains the interaction data set between equipment;Breeding equipment is regarded as node, and dependency graph model is constructed;At least one breeding equipment is imported simulation fault signal, adopts bayesian network to infer fault propagation probability, obtains a chain reaction path;When the activation value of node in chain reaction path exceeds preset activation threshold, according to path length and the influence weight of path to dependency graph model, the overall reliability level is calculated;Reliability level is compared and analyzed with the reliability data of each fault record sample in historical operation log data;When similar fault scene is matched, it is determined that the overall operating state exists anomaly, and corresponding breeding equipment exists anomaly, and the application can effectively reduce the hysteresis of prior art breeding equipment reliability evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of device state recognition, and in particular to a marine ranching device reliability evaluation method and system based on a large model. BACKGROUND

[0002] Marine ranching is an important pillar of modern marine economic development, and through the construction of an artificial ecosystem, it realizes the sustainable use of marine resources and has important significance for food security and ecological balance. However, a number of breeding devices are often required in marine ranching, including but not limited to oxygen supply devices, bait feeding devices, and environmental monitoring devices. The reliability of these breeding devices will directly affect production efficiency and economic benefits.

[0003] Currently, most breeding device reliability evaluations focus on single device performance optimization or static system design, ignoring the dynamic interdependence between devices, which can easily overlook the chain reaction caused by device failure, resulting in a lag in reliability evaluation. SUMMARY

[0004] The present application provides a marine ranching device reliability evaluation method and system based on a large model to solve the technical problem of how to reduce the lag in marine ranching device reliability evaluation.

[0005] To solve the above technical problems, the present application provides a marine ranching device reliability evaluation method based on a large model, wherein the marine ranching includes a plurality of types of breeding devices;

[0006] The reliability evaluation method comprises:

[0007] Obtaining the operating parameters of each breeding device in the marine ranching through a pre-set sensor network; and analyzing the operating parameters to obtain an inter-device interaction data set;

[0008] Based on the inter-device interaction data set, the breeding devices are taken as nodes to construct a dependency graph model; wherein a large model is used in the construction process of the dependency graph model, and the parameter quantity of the large model is greater than a pre-set quantity threshold;

[0009] Based on the dependency graph model, at least one breeding device is input with a simulated fault signal, and a Bayesian network is used to infer the fault propagation probability to obtain at least one chain reaction path; the chain reaction path includes a plurality of nodes;

[0010] monitoring the chain reaction path; and when the activation value of at least one node in the chain reaction path exceeds a preset activation threshold, calculating a reliability level of the entire aquaculture equipment according to the path length of the chain reaction path and the influence weight of the chain reaction path on the dependency graph model;

[0011] obtaining historical operation log data; and comparing and analyzing the reliability level with the reliability data of each failure record sample in the historical operation log data;

[0012] According to the comparison and analysis result, when a similar failure scenario is matched in the historical operation log data, it is determined that the operation state of the entire aquaculture equipment in the marine ranching exists abnormity, and it is determined that the operation state of the aquaculture equipment corresponding to the node whose activation value exceeds the preset activation threshold exists abnormity, thereby realizing the reliability evaluation of the operation of the aquaculture equipment.

[0013] As a preferred solution, the interaction data set includes interaction frequency and interaction times in a preset time period; and the dependency graph model is constructed by taking the aquaculture equipment as nodes based on the interaction data set between the equipment, including:

[0014] The aquaculture equipment is taken as nodes, the interaction frequency is taken as edges, and the interaction times in the preset time period after normalization processing are taken as weights to construct an initial graph model;

[0015] A directed graph is created based on the initial graph model, and the node features of the directed graph are initialized as the computing capacity of the aquaculture equipment to obtain an initialized directed graph; the computing capacity is represented by CPU frequency;

[0016] According to the edge weight of the initialized directed graph, the dynamic dependency strength between the equipment is obtained;

[0017] The dynamic dependency strength between the equipment is aggregated by a pre-trained graph neural network to capture the dependency relationship between the equipment; wherein the aggregation function of the aggregation processing is mean aggregation, the dimension number of the hidden layer of the pre-trained graph neural network is 64, the learning rate of the pre-trained graph neural network is 0.01, the pre-trained graph neural network adopts Adam optimizer, and the loss function of the pre-trained graph neural network adopts mean square error;

[0018] The dependency graph model is obtained based on the output of the pre-trained graph neural network.

[0019] As a preferred solution, the type of the aquaculture equipment includes oxygen supply equipment; and the dependency graph model is used to import a simulated failure signal to at least one aquaculture equipment, a Bayesian network is used to infer a failure propagation probability to obtain at least one chain reaction path, including:

[0020] The dependency graph model is segmented based on the oxygen supply equipment corresponding node by using a depth-first algorithm, to obtain a subgraph associated with the oxygen supply equipment, and a subgraph node set is obtained;

[0021] A dependency relationship matrix between nodes is constructed according to the nodes in the subgraph node set; the dependency relationship node matrix is represented by an adjacency matrix;

[0022] An analog fault signal is injected into the oxygen supply equipment corresponding node to obtain an initial fault state;

[0023] Based on the dependency relationship matrix and the initial fault state, a Bayesian network is used to infer a fault propagation probability distribution;

[0024] At least one chain reaction path is obtained by traversing the subgraph node set according to the fault propagation probability distribution.

[0025] As a preferred solution, the historical operation log data is obtained; the reliability level is compared and analyzed with the reliability data of each fault record sample in the historical operation log data, including:

[0026] The historical operation log data is obtained;

[0027] A plurality of fault record samples are extracted from the historical operation log data; wherein the sample information of each fault record sample includes a fault start time, a device ID, a fault type and a duration;

[0028] The fault probability of the sample device in the mariculture farm is calculated according to the sample information by a fault tree analysis algorithm;

[0029] The key device is determined based on the fault probability of the sample device;

[0030] The reliability data of the mariculture farm as a whole is calculated through the key device;

[0031] According to the reliability data, each fault record sample is clustered to obtain a plurality of fault scenarios;

[0032] The reliability level is compared and analyzed with the reliability data of each fault record sample in the historical operation log data;

[0033] According to the results of the comparative analysis, the fault scenario with the highest matching degree and the matching degree greater than the preset matching threshold is determined as the similar fault scenario.

[0034] As a preferred solution, the operation parameters of each breeding device in the mariculture farm are obtained through a preset sensor network, including:

[0035] The dissolved oxygen content of the water in the area where each aquaculture device in the mariculture farm is located is obtained through a dissolved oxygen sensor, and oxygen content data is obtained.

[0036] The motor load current of each aquaculture device in the mariculture farm is obtained through a current sensor, and device load data is obtained.

[0037] The water temperature data in the area where each aquaculture device in the mariculture farm is located is obtained through a temperature sensor.

[0038] The operating parameters are obtained according to the water temperature data, oxygen content data and device load data.

[0039] As a preferred solution, the analysis according to the operating parameters to obtain the device interaction data set comprises:

[0040] A data packet is generated based on the operating parameters through a preset transmission protocol, and the data packet is transmitted to an edge computing node; wherein the format of the data packet is JSON format, and the data packet further contains the device ID, timestamp and parameter value corresponding to the operating parameters;

[0041] The edge computing node is controlled to aggregate the data packet to a cloud database using the MQTT protocol, and the data packet is processed as time series data in the cloud database;

[0042] The data packet is associated and analyzed through a preset association analysis model to obtain an association analysis result; and the Pearson correlation coefficient between the water temperature data, oxygen content data and device load data is calculated based on the data packet;

[0043] The device interaction data set is analyzed according to the association analysis result and the Pearson correlation coefficient.

[0044] As a preferred solution, before the data packet is associated and analyzed through the preset association analysis model to obtain the association analysis result, it further comprises: verifying the data packet through a CRC32 verification algorithm, and obtaining the verified data packet when the verification is passed; if the verification fails, the edge computing node is controlled to aggregate the data packet to the cloud database using the MQTT protocol again.

[0045] Correspondingly, the present application also provides a mariculture farm aquaculture device reliability evaluation system based on a large model, wherein the mariculture farm comprises a plurality of types of aquaculture devices.

[0046] The reliability evaluation system comprises an analysis module, a model construction module, an inference module, a calculation module, a comparison module and an evaluation module.

[0047] The analysis module is configured to acquire operation parameters of each aquaculture device in the mariculture through a preset sensor network; and analyze the operation parameters to obtain an interaction dataset between devices;

[0048] The model construction module is configured to construct a dependency graph model based on the interaction dataset between devices, taking the aquaculture devices as nodes; wherein a large model is used in the process of constructing the dependency graph model, and a parameter quantity of the large model is greater than a preset quantity threshold;

[0049] The inference module is configured to input a simulated fault signal to at least one aquaculture device based on the dependency graph model, infer a fault propagation probability by using a Bayesian network, and obtain at least one chain reaction path; the chain reaction path includes multiple nodes;

[0050] The calculation module is configured to monitor the chain reaction path; and when an activation value of at least one node in the chain reaction path exceeds a preset activation threshold, calculate a reliability level of all aquaculture devices as a whole according to a path length of the chain reaction path and an influence weight of the chain reaction path on the dependency graph model;

[0051] The comparison module is configured to acquire historical operation log data; and compare and analyze the reliability level with reliability data of each fault record sample in the historical operation log data;

[0052] The evaluation module is configured to determine that an operation state of the aquaculture devices in the mariculture as a whole is abnormal, and determine that an operation state of an aquaculture device corresponding to a node whose activation value exceeds the preset activation threshold is abnormal, according to a comparison and analysis result, when a similar fault scenario is matched in the historical operation log data, thereby realizing reliability evaluation of the operation of the aquaculture devices.

[0053] As a preferred solution, the interaction dataset includes an interaction frequency and an interaction number in a preset time period; the model construction module constructs a dependency graph model based on the interaction dataset between devices, taking the aquaculture devices as nodes, including:

[0054] The model construction module takes the aquaculture devices as nodes, takes the interaction frequency as an edge, and takes the interaction number in the preset time period after normalization processing as a weight to construct an initial graph model;

[0055] A directed graph is created based on the initial graph model, and a node feature of the directed graph is initialized as a computing capacity of the aquaculture device to obtain an initialized directed graph; the computing capacity is represented by a CPU frequency;

[0056] A dynamic dependency strength between devices is obtained according to an edge weight of the initialized directed graph;

[0057] The dependence strength of the device inter-dynamics is aggregated by a pre-trained graph neural network to capture the dependence relationship between devices, wherein an aggregation function of the aggregation processing is mean aggregation, a dimension number of a hidden layer of the pre-trained graph neural network is 64, a learning rate of the pre-trained graph neural network is 0.01, the pre-trained graph neural network adopts an Adam optimizer, and a loss function of the pre-trained graph neural network adopts a mean square error;

[0058] Based on the output of the pre-trained graph neural network, the dependence graph model is obtained.

[0059] As a preferred solution, the type of the breeding device includes an oxygen supply device; the inference module injects a simulated fault signal into at least one breeding device based on the dependence graph model, infers a fault propagation probability using a Bayesian network, and obtains at least one chain reaction path, including:

[0060] The inference module performs segmentation processing on the dependence graph model based on the node corresponding to the oxygen supply device to obtain a subgraph associated with the oxygen supply device, and obtains a subgraph node set;

[0061] A dependence relationship matrix between nodes is constructed according to the nodes in the subgraph node set; the dependence relationship node matrix is represented by an adjacency matrix;

[0062] A simulated fault signal is injected into the node corresponding to the oxygen supply device to obtain an initial fault state;

[0063] Based on the dependence relationship matrix and the initial fault state, a Bayesian network is used to infer a fault propagation probability distribution;

[0064] According to the fault propagation probability distribution, the subgraph node set is traversed to obtain at least one chain reaction path.

[0065] As a preferred solution, the comparison module obtains historical operation log data; the reliability level is compared and analyzed with reliability data of each fault record sample in the historical operation log data, including:

[0066] The comparison module obtains historical operation log data;

[0067] A plurality of fault record samples are extracted from the historical operation log data; wherein sample information of each fault record sample includes a fault start time, a device ID, a fault type, and a duration;

[0068] A fault probability of a sample device in the mariculture farm is calculated according to the sample information by a fault tree analysis algorithm;

[0069] A key device is determined based on the fault probability of the sample device;

[0070] reliability data of the whole mariculture farm is calculated by the key equipment;

[0071] According to the reliability data, each failure record sample is clustered to obtain a plurality of failure scenarios;

[0072] The reliability level is compared and analyzed with the reliability data of each failure record sample in the historical operation log data;

[0073] According to the result of the comparative analysis, the failure scenario with the highest matching degree and the matching degree greater than the preset matching threshold is determined as the similar failure scenario.

[0074] As a preferred solution, the analysis module obtains the operation parameters of each breeding equipment in the mariculture farm through a preset sensor network, including:

[0075] The analysis module obtains the dissolved oxygen content of the water in the area where each breeding equipment in the mariculture farm is located through a dissolved oxygen sensor to obtain oxygen content data;

[0076] Through a current sensor, the motor load current of each breeding equipment in the mariculture farm is obtained to obtain equipment load data;

[0077] The water temperature data of the area where each breeding equipment in the mariculture farm is located is obtained through a temperature sensor;

[0078] According to the water temperature data, oxygen content data and equipment load data, the operation parameters are obtained.

[0079] As a preferred solution, the analysis module analyzes the operation parameters to obtain an inter-equipment interaction data set, including:

[0080] The analysis module generates a data packet based on the operation parameters through a preset transmission protocol, and transmits the data packet to an edge computing node; wherein the format of the data packet is JSON format, and the data packet further contains the device ID, timestamp and parameter value corresponding to the operation parameters;

[0081] The edge computing node is controlled to use the MQTT protocol to aggregate the data packet to a cloud database, and the data packet is processed as time series data in the cloud database;

[0082] The verified data packet is associated analyzed through a preset association analysis model to obtain an association analysis result; and the Pearson correlation coefficient between the water temperature data, oxygen content data and equipment load data is calculated based on the data packet;

[0083] According to the association analysis result and the Pearson correlation coefficient, the inter-equipment interaction data set is analyzed and obtained.

[0084] As a preferred solution, the reliability evaluation system further comprises a verification module, configured to, before the analysis module performs correlation analysis on the data packet by using a preset correlation analysis model to obtain a correlation analysis result: verify the data packet by using a CRC32 verification algorithm, and when the verification is passed, obtain the verified data packet; and when the verification fails, re-control the edge computing node to aggregate the data packet to a cloud database by using an MQTT protocol.

[0085] Compared with the prior art, the application has the following beneficial effects:

[0086] The application provides a method and system for evaluating the reliability of marine ranching equipment based on a large model. The marine ranching environment contains multiple types of equipment. The method includes obtaining operating parameters of each piece of equipment in the marine ranching environment through a pre-set sensor network, analyzing the operating parameters to obtain an interaction dataset between the equipment, constructing a dependency graph model based on the interaction dataset, using a large model with a parameter quantity greater than a pre-set threshold in the construction process of the dependency graph model, inputting a simulated fault signal into at least one piece of equipment based on the dependency graph model, inferring the fault propagation probability using a Bayesian network, obtaining at least one chain reaction path, and monitoring the chain reaction path. The chain reaction path contains multiple nodes. When the activation value of at least one node in the chain reaction path exceeds a pre-set activation threshold, the reliability level of all equipment is calculated based on the path length of the chain reaction path and the influence weight of the chain reaction path on the dependency graph model. The reliability level is compared with the reliability data of each fault record sample in the historical operation log data. When a similar fault scenario is matched in the historical operation log data, it is determined that the operation state of the equipment in the marine ranching environment is abnormal, and the operation state of the equipment corresponding to the node with an activation value exceeding the pre-set activation threshold is also abnormal, thereby achieving reliability evaluation of the equipment operation. The application constructs a dependency graph model based on a large model, treats the equipment in the marine ranching environment as nodes that are related to each other, inputs a simulated fault signal, and uses a Bayesian network to infer the fault propagation probability and the chain reaction path. This directly solves the defect of ignoring the dynamic interdependence between devices in the prior art. By monitoring the activation value of the node in the chain reaction path in real time and dynamically calculating the overall reliability level of all equipment, the method can capture and quantify the potential propagation effect of the fault in the device network in advance, significantly reduce the lag of reliability evaluation, and make the evaluation results more realistic and dynamic in the complex marine ranching environment. In addition, by actively inputting a simulated fault signal, the method realizes forward-looking fault scenario deduction, can not only find potential risk points (nodes with activation values exceeding the threshold) in the current state, but also systematically reveals the chain reaction path and its impact range when a fault occurs, providing a strong decision basis for proactive preventive maintenance and risk control. By comparing the real-time calculated reliability level with the historical operation log data, when a similar fault scenario is matched, not only is the overall operation state of all equipment abnormal, but also the specific source equipment (the equipment corresponding to the node with an activation value exceeding the threshold) that caused the abnormality can be accurately located. This intelligent diagnosis based on historical experience greatly improves the accuracy and reliability of fault identification and state determination, and reduces false positives and false negatives. BRIEF DESCRIPTION OF DRAWINGS

[0087] Figure 1 FIG. 1 is a flowchart of an embodiment of the method for evaluating reliability of marine ranching equipment based on a large model provided by the present application.

[0088] Figure 2 FIG. 2 is a flowchart of preferred implementation one of the method for evaluating reliability of marine ranching equipment based on a large model provided by the present application.

[0089] Figure 3 FIG. 3 is a flowchart of preferred implementation two of the method for evaluating reliability of marine ranching equipment based on a large model provided by the present application.

[0090] Figure 4 FIG. 4 is a flowchart of preferred implementation three of the method for evaluating reliability of marine ranching equipment based on a large model provided by the present application.

[0091] Figure 5 FIG. 5 is a structural diagram of an embodiment of the system for evaluating reliability of marine ranching equipment based on a large model provided by the present application. DETAILED DESCRIPTION

[0092] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only 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 skilled in the art without creative labor fall within the scope of protection of the present application.

[0093] Embodiment one

[0094] Please refer to Figure 1 , Figure 1 The method for evaluating reliability of marine ranching equipment based on a large model provided by the present application comprises a plurality of types of marine ranching equipment.

[0095] The method for evaluating reliability of marine ranching equipment based on a large model comprises steps S101 to S106; each step is described as follows:

[0096] In step S101, the running parameters of each marine ranching equipment in the marine ranching are obtained through a preset sensor network; and the interaction data set between the equipment is obtained by analyzing the running parameters.

[0097] In the embodiment, the preset sensor network includes multiple types of sensors (such as dissolved oxygen sensors, temperature sensors, and current sensors, etc.), and each type of sensor has one or more quantity, and each sensor is arranged at different positions in the mariculture farm, thereby forming the above-mentioned preset sensor network.

[0098] The cultivation equipment can also include multiple types, such as oxygen supply equipment, bait feeding device, and environmental monitoring equipment, etc. The above-mentioned preset sensor network is used to collect parameters of the environment or the running state of the cultivation equipment, and the above-mentioned running parameters are obtained.

[0099] Preferably, as shown in the step S101, the running parameters of each cultivation equipment in the mariculture farm are obtained through the preset sensor network, including steps S201 to S204, and each step is described as follows: Figure 2

[0100] The step S201 obtains the dissolved oxygen content of the water in the area where each cultivation equipment in the mariculture farm is located through the dissolved oxygen sensor, and obtains the oxygen content data;

[0101] The step S202 obtains the motor load current of each cultivation equipment in the mariculture farm through the current sensor, and obtains the equipment load data;

[0102] The step S203 obtains the water temperature data of the area where each cultivation equipment in the mariculture farm is located through the temperature sensor;

[0103] The step S204 obtains the running parameters according to the water temperature data, the oxygen content data, and the equipment load data.

[0104] On the basis of obtaining the running parameters through the preset sensor network in the step S101, the preferred embodiment realizes the data fusion of the dissolved oxygen, current, and temperature three types of sensors. The existing monitoring scheme is often independent: the dissolved oxygen meter only looks at the dissolved oxygen, and the ammeter only looks at the power consumption, lacking a unified space-time reference. The present scheme synchronously collects the water dissolved oxygen, the motor load current, and the water temperature, and forms a three-dimensional running parameter vector under the same time stamp, so that the subsequent calculation of the edge weight (interaction intensity) in the graph model has the comparability across the physical domain; at the same time, the water temperature as an environmental coupling factor is included, and the drift error of the dissolved oxygen and the load current can be corrected in real time. Compared with the prior art, the fusion strategy upgrades the originally fragmented monitoring data to the coupling data of the environment-equipment, provides more real and more robust input for the subsequent graph model, and significantly reduces the misjudgment caused by the distortion of single-point data.

[0105] Further, the dissolved oxygen sensor can adopt a DO-100 dissolved oxygen sensor, and samples once per second, and the temperature sensor can adopt a DS18B20 model.

[0106] ​Further, in order to analyze the correlation between the water temperature data, oxygen content data and equipment load data of different types of operating parameters, as shown in Figure 3 According to the operating parameters, the step S101 analyzes to obtain the interaction data set between devices, including steps S301 to S304, wherein each step is described as follows:

[0107] In step S301, a data packet is generated based on the operating parameters through a preset transmission protocol, and the data packet is transmitted to an edge computing node; wherein the format of the data packet is JSON format, and the data packet further contains the device ID, timestamp and parameter value corresponding to the operating parameters;

[0108] In step S302, the edge computing node is controlled to use the MQTT protocol to aggregate the data packet to a cloud database, and the data packet is processed as time series data in the cloud database;

[0109] In step S303, the data packet is analyzed by a preset correlation analysis model to obtain a correlation analysis result, and the Pearson correlation coefficient between the water temperature data, oxygen content data and equipment load data is calculated based on the data packet;

[0110] In step S304, the interaction data set between devices is analyzed based on the correlation analysis result and the Pearson correlation coefficient.

[0111] In addition, in some examples, in order to ensure data integrity, before the step S303 analyzes the data packet by the preset correlation analysis model to obtain the correlation analysis result, it further includes: verifying the data packet by a CRC32 verification algorithm, and obtaining the verified data packet when the verification is passed; when the verification fails, the edge computing node is controlled to use the MQTT protocol to aggregate the data packet to the cloud database again.

[0112] The preferred embodiment in Figure 2On the basis of the three-dimensional parameters of the illustrated embodiment, the whole link of data packet generation, transmission, verification and correlation analysis is further standardized: JSON+MQTT+CRC32+Pearson correlation coefficient. The existing marine ranching system often has packet loss, out-of-order or dirty data due to inconsistent transmission protocols and missing verification, resulting in distorted subsequent correlation analysis. The present scheme ensures data integrity through CRC32, guarantees high arrival rate under low bandwidth through MQTT, and quantifies the linear correlation of dissolved oxygen-load current-water temperature as the "strength" input of the interaction dataset between devices. Compared with the prior art, the link standardizes and automates the conversion process from "sensor raw value to reliable time series to interaction strength", significantly reducing the workload of manual data cleaning, improving the accuracy of edge weight of the dependency graph model, and further improving the reliability of the overall reliability evaluation.

[0113] In step S102, based on the interaction dataset between devices, the aquaculture equipment is taken as a node to construct a dependency graph model.

[0114] In the present embodiment, a large model can be used in the process of constructing the dependency graph model, and the parameter quantity of the large model is greater than a preset quantity threshold. According to the prior art, it can be understood that the artificial intelligence large model is a new concept proposed in recent years. It is usually pre-trained on massive data through self-supervised learning or semi-supervised learning, and then its performance and ability are further optimized through instruction fine-tuning and alignment. The large model has the characteristics of large parameter quantity, large training data and large computing resources, and has the ability to solve general tasks, follow human instructions and perform complex reasoning. The main categories of artificial intelligence large models include large language models, visual large models, multi-modal large models and basic scientific large models, etc.

[0115] In some preferred embodiments, for the construction mode of the dependency graph model, the interaction dataset includes interaction frequency and interaction times in a preset time period (for example, assuming that the dataset contains 100 farming devices, each farming device records the interaction frequency with other farming devices within 24 hours, such as that farming device A interacts with farming device B for 10 times, farming device C interacts with farming device D for 5 times, and the like); and the construction of the dependency graph model based on the device interaction dataset, taking the farming devices as nodes, includes: constructing an initial graph model by taking the farming devices as nodes, the interaction frequency as edges, and the normalized interaction times in the preset time period as weights; creating a directed graph based on the initial graph model using the NetworkX library, and initializing the node features of the directed graph as the computing power of the farming devices to obtain an initialized directed graph; the computing power is represented by CPU frequency (for example, farming device A is 2.5 GHz, and farming device B is 3.0 GHz); obtaining the dynamic dependency strength between devices according to the edge weights of the initialized directed graph; and performing aggregation processing on the dynamic dependency strength between devices by using a pre-trained graph neural network to capture the dependency relationship between devices; wherein the aggregation function of the aggregation processing is mean aggregation, the number of dimensions of the hidden layer of the pre-trained graph neural network is 64, the learning rate of the pre-trained graph neural network is 0.01, the pre-trained graph neural network uses the Adam optimizer, and the loss function of the pre-trained graph neural network uses the mean square error; and obtaining the dependency graph model based on the output of the pre-trained graph neural network.

[0116] Preferably, in the pre-training process of the graph neural network, the graph neural network can be trained for 100 epochs, the sample dataset can be divided into 80% training set and 20% test set, and the R 2 The score reaches 0.85 to determine that the convergence condition is met, which indicates that the model can effectively capture the dependency relationship.

[0117] In step S103, based on the dependency graph model, at least one farming device is inputted with a simulated fault signal, a Bayesian network is used to infer a fault propagation probability, and at least one chain reaction path is obtained.

[0118] In this embodiment, the simulated fault signal is used to determine the possible chain reaction when a fault occurs by simulating and assuming the fault. The chain reaction path includes multiple nodes.

[0119] In some preferred embodiments, the inference of the fault propagation probability and the chain reaction path can be inferred from the oxygen supply device, so that Figure 4As shown, step S103 is based on the dependency graph model, at least one breeding device is injected into the simulation fault signal, the Bayesian network is used to infer the fault propagation probability, and at least one chain reaction path is obtained, including steps S401 to S405, wherein each step is described as follows:

[0120] Step S401, based on the oxygen supply equipment corresponding node, the dependency graph model is segmented and processed to obtain a subgraph associated with the oxygen supply equipment, and a subgraph node set is obtained;

[0121] Step S402, according to the nodes in the subgraph node set, a dependency relationship matrix between nodes is constructed; the dependency relationship node matrix is represented by an adjacency matrix;

[0122] Step S403, injecting a simulation fault signal into the oxygen supply equipment corresponding node to obtain a fault initial state;

[0123] Step S404, based on the dependency relationship matrix and the fault initial state, a Bayesian network is used to infer the fault propagation probability distribution;

[0124] Step S405, according to the fault propagation probability distribution, the subgraph node set is traversed to obtain at least one chain reaction path.

[0125] The preferred embodiment takes the oxygen supply equipment as the core link, first performs subgraph segmentation on the dependency graph model by using the depth-first algorithm, then constructs an adjacency matrix and infers the fault propagation by using the Bayesian network. This approach reduces the dimensionality of the high-dimensional full graph problem to the oxygen supply equipment associated subgraph, significantly reducing the computational complexity of Bayesian inference; at the same time, the adjacency matrix explicitly quantifies the dependency relationship, making the oxygen supply equipment fault propagation probability calculation in the subgraph more accurate and faster. Compared with the existing technology which still uses single reliability index or simple redundancy design for oxygen supply equipment, the present scheme first realizes the progressive fault deduction of "oxygen supply equipment-subgraph-full system-single device or system overall evaluation", which not only avoids the exponential explosion of full graph calculation, but also preserves the integrity of the chain reaction path, thereby further improving the accuracy of oxygen supply system reliability evaluation under the premise of ensuring real-time.

[0126] Step S104, monitoring the chain reaction path; and when the activation value of at least one node in the chain reaction path exceeds a preset activation threshold, calculating the reliability level of all breeding devices according to the path length of the chain reaction path and the influence weight of the chain reaction path on the dependency graph model.

[0127] In the embodiment, the monitoring of the cascading reaction path can specifically be monitoring activation values of multiple nodes in the cascading reaction path. When the activation value of at least one node in the cascading reaction path exceeds a preset activation threshold, it can be taken as a trigger condition for the calculation of the overall reliability level of the aquaculture equipment, and the overall reliability level of the aquaculture equipment is calculated by the path length of the cascading reaction path (for example, represented by the distance between the start node and the end node, the number of nodes between the start node and the end node, etc.) and the influence weight of the cascading reaction path on the dependency graph model.

[0128] In step S105, historical operation log data is obtained, and the reliability level is compared and analyzed with reliability data of each failure record sample in the historical operation log data.

[0129] In the embodiment, the reliability level is compared and analyzed with reliability data of each failure record sample in the historical operation log data, so as to determine whether the operation state of the overall aquaculture equipment in the mariculture farm is abnormal, or whether the operation state of the aquaculture equipment corresponding to the node with the activation value exceeding the preset activation threshold is abnormal.

[0130] In some embodiments, the historical operation log data is obtained, and the reliability level is compared and analyzed with reliability data of each failure record sample in the historical operation log data, including:

[0131] The historical operation log data is obtained, and multiple failure record samples are extracted from the historical operation log data. The sample information of each failure record sample includes a failure start time, a device ID, a failure type, and a duration. A fault tree analysis algorithm is used to calculate a failure probability of a sample device in the mariculture farm according to the sample information. A key device is determined based on the failure probability of the sample device. The reliability data of the overall mariculture farm is calculated by the key device. Each failure record sample is clustered according to the reliability data, and multiple failure scenarios are obtained. The reliability level is compared and analyzed with the reliability data of each failure record sample in the historical operation log data. According to the comparison and analysis result, a failure scenario with the highest matching degree and a matching degree greater than a preset matching threshold is determined as a similar failure scenario.

[0132] The embodiment introduces a combined process of fault tree analysis (FTA), key device identification and reliability clustering. Compared with the prior art of only simple statistics or threshold alarm on logs, the embodiment quantifies the fault starting time, device ID, fault type and duration of each fault record as the fault probability of the sample device through fault tree analysis, and then reversely locates the key device and clusters to generate typical fault scenarios. This process converts discrete and fragmented log data into reusable "fault scenario library". When the real-time reliability level needs to be matched with the historical scenario, the matching speed is improved by an order of magnitude in the clustered scenario cluster. At the same time, since the key device has been calibrated in advance, the system can immediately give the abnormal attribution chain from the key device to the whole system after successful matching, significantly reducing false negatives and false positives.

[0133] In step S106, according to the comparison and analysis result, when a similar fault scenario is matched in the historical operation log data, it is determined that the operation state of the whole aquaculture device in the marine ranching is abnormal, and the operation state of the aquaculture device corresponding to the node with the activation value exceeding the preset activation threshold is also determined to be abnormal, thereby realizing the reliability evaluation of the operation of the aquaculture device.

[0134] According to the comparison and analysis result, when a similar fault scenario is matched, it is determined that the operation state of the whole aquaculture device in the marine ranching is abnormal, and the related technical personnel need to eliminate the possible risks in the marine ranching or eliminate the situation affecting the aquaculture efficiency. At the same time, it is also determined that the operation state of the aquaculture device corresponding to the node with the activation value exceeding the preset activation threshold is abnormal. It needs to be emphasized that when the node activation value exceeds the preset activation threshold, the aquaculture device cannot be determined as an abnormal operation state. Only when the node activation value exceeds the preset activation threshold and a similar fault scenario is matched, it can be determined to be abnormal. In this way, the situation of normal operation fluctuation can be effectively eliminated, thereby avoiding misjudgment.

[0135] Correspondingly, as shown in Figure 5 The present application also provides a marine ranching aquaculture device reliability evaluation system 500 based on a large model. The marine ranching includes multiple types of aquaculture devices.

[0136] The reliability evaluation system 500 includes an analysis module 501, a model construction module 502, an inference module 503, a calculation module 504, a comparison module 505 and an evaluation module 506.

[0137] The analysis module 501 is used to obtain the operation parameters of each aquaculture device in the marine ranching through a preset sensor network; and analyze the operation parameters to obtain an interaction data set between devices.

[0138] The model construction module 502 is configured to construct a dependency graph model by taking the aquaculture devices as nodes based on the inter-device interaction data set; wherein a large model is used in the process of constructing the dependency graph model, and a parameter quantity of the large model is greater than a preset quantity threshold;

[0139] The inference module 503 is configured to input a simulated fault signal into at least one aquaculture device based on the dependency graph model, infer a fault propagation probability by using a Bayesian network, and obtain at least one chain reaction path; the chain reaction path includes a plurality of nodes;

[0140] The calculation module 504 is configured to monitor the chain reaction path, and when an activation value of at least one node in the chain reaction path exceeds a preset activation threshold, calculate a reliability level of all aquaculture devices as a whole according to a path length of the chain reaction path and an influence weight of the chain reaction path on the dependency graph model;

[0141] The comparison module 505 is configured to obtain historical operation log data, and compare and analyze the reliability level with reliability data of each fault record sample in the historical operation log data;

[0142] The evaluation module 506 is configured to determine that an operation state of the aquaculture devices in the marine ranching as a whole is abnormal, and determine that an operation state of an aquaculture device corresponding to a node whose activation value exceeds the preset activation threshold is abnormal, when a similar fault scenario is matched in the historical operation log data, thereby achieving reliability evaluation of the operation of the aquaculture devices.

[0143] As a preferred solution, the interaction data set includes an interaction frequency and an interaction frequency within a preset time period; the model construction module 502 constructs a dependency graph model by taking the aquaculture devices as nodes based on the inter-device interaction data set, including:

[0144] The model construction module 502 constructs an initial graph model by taking the aquaculture devices as nodes, taking the interaction frequency as an edge, and taking the interaction frequency within the preset time period after normalization processing as a weight;

[0145] A directed graph is created based on the initial graph model, and a node feature of the directed graph is initialized as a computing capacity of the aquaculture device, to obtain an initialized directed graph; the computing capacity is represented by a CPU frequency;

[0146] A dynamic dependency strength between devices is obtained according to an edge weight of the initialized directed graph;

[0147] The dependence strength of the device inter-dynamics is aggregated by a pre-trained graph neural network to capture the dependence relationship between devices, wherein an aggregation function of the aggregation processing is mean aggregation, a dimension number of a hidden layer of the pre-trained graph neural network is 64, a learning rate of the pre-trained graph neural network is 0.01, the pre-trained graph neural network adopts an Adam optimizer, and a loss function of the pre-trained graph neural network adopts a mean square error;

[0148] Based on the output of the pre-trained graph neural network, the dependence graph model is obtained.

[0149] As a preferred solution, the type of the breeding device includes an oxygen supply device; the inference module 503 injects a simulated fault signal into at least one breeding device based on the dependence graph model, infers a fault propagation probability using a Bayesian network, and obtains at least one chain reaction path, including:

[0150] The inference module 503 performs segmentation processing on the dependence graph model based on the node corresponding to the oxygen supply device to obtain a subgraph associated with the oxygen supply device, and obtains a subgraph node set;

[0151] A dependence relationship matrix between nodes is constructed according to the nodes in the subgraph node set; the dependence relationship node matrix is represented by an adjacency matrix;

[0152] A simulated fault signal is injected into the node corresponding to the oxygen supply device to obtain an initial fault state;

[0153] Based on the dependence relationship matrix and the initial fault state, a Bayesian network is used to infer a fault propagation probability distribution;

[0154] According to the fault propagation probability distribution, the subgraph node set is traversed to obtain at least one chain reaction path.

[0155] As a preferred solution, the comparison module 505 obtains historical operation log data; the reliability level is compared and analyzed with reliability data of each fault record sample in the historical operation log data, including:

[0156] The comparison module 505 obtains historical operation log data;

[0157] A plurality of fault record samples are extracted from the historical operation log data; wherein sample information of each fault record sample includes a fault start time, a device ID, a fault type, and a duration;

[0158] The fault probability of the sample device in the mariculture farm is calculated according to the sample information by a fault tree analysis algorithm;

[0159] Based on the fault probability of the sample device, a key device is determined.

[0160] calculate reliability data of the whole mariculture farm through the key equipment;

[0161] cluster each failure record sample according to the reliability data to obtain a plurality of failure scenarios;

[0162] compare and analyze the reliability level with the reliability data of each failure record sample in the historical operation log data;

[0163] determine, according to the result of the comparison and analysis, a failure scenario with the highest matching degree and a matching degree greater than a preset matching threshold as a similar failure scenario.

[0164] As a preferred solution, the analysis module 501 obtains the operation parameters of each breeding equipment in the mariculture farm through a preset sensor network, including:

[0165] The analysis module 501 obtains the dissolved oxygen content of the water in the area where each breeding equipment in the mariculture farm is located through a dissolved oxygen sensor to obtain oxygen content data;

[0166] obtain the motor load current of each breeding equipment in the mariculture farm through a current sensor to obtain equipment load data;

[0167] obtain the water temperature data of the area where each breeding equipment in the mariculture farm is located through a temperature sensor;

[0168] obtain the operation parameters according to the water temperature data, the oxygen content data and the equipment load data.

[0169] As a preferred solution, the analysis module 501 analyzes according to the operation parameters to obtain an inter-equipment interaction data set, including:

[0170] The analysis module 501 generates a data packet based on the operation parameters through a preset transmission protocol and transmits the data packet to an edge computing node; wherein the format of the data packet is JSON format, and the data packet further contains the device ID, timestamp and parameter value corresponding to the operation parameters;

[0171] control the edge computing node to aggregate the data packet to a cloud database using the MQTT protocol, and process the data packet to time series data in the cloud database;

[0172] perform correlation analysis on the verified data packet through a preset correlation analysis model to obtain a correlation analysis result; and calculate the Pearson correlation coefficient between the water temperature data, the oxygen content data and the equipment load data based on the data packet;

[0173] obtain the inter-equipment interaction data set according to the correlation analysis result and the Pearson correlation coefficient.

[0174] As a preferred solution, the reliability evaluation system 500 further comprises a verification module, which is configured to, before the analysis module 501 performs correlation analysis on the data packet by using a preset correlation analysis model to obtain a correlation analysis result: verify the data packet by using a CRC32 verification algorithm, and when the verification is passed, obtain the verified data packet; and when the verification fails, re-control the edge computing node to aggregate the data packet to a cloud database by using an MQTT protocol.

[0175] Compared with the prior art, the application has the following beneficial effects:

[0176] The application provides a large model-based reliability evaluation method and system for mariculture equipment in a marine ranching area. The marine ranching area contains multiple types of mariculture equipment. The method includes: obtaining operation parameters of each mariculture equipment in the marine ranching area through a preset sensor network; analyzing the operation parameters to obtain an interaction dataset between the equipment; constructing a dependency graph model based on the interaction dataset between the equipment, taking the mariculture equipment as nodes; in the process of constructing the dependency graph model, a large model is used, and the parameter quantity of the large model is greater than a preset quantity threshold; based on the dependency graph model, at least one mariculture equipment is input with a simulated fault signal, a fault propagation probability is inferred using a Bayesian network, and at least one chain reaction path is obtained; the chain reaction path contains multiple nodes; the chain reaction path is monitored; when the activation value of at least one node in the chain reaction path exceeds a preset activation threshold, the reliability level of all mariculture equipment is calculated according to the path length of the chain reaction path and the influence weight of the chain reaction path on the dependency graph model; historical operation log data is obtained; the reliability level is compared and analyzed with the reliability data of each fault record sample in the historical operation log data; when a similar fault scenario is matched in the historical operation log data according to the comparison and analysis result, it is determined that the operation state of the mariculture equipment in the marine ranching area as a whole is abnormal, and it is also determined that the operation state of the mariculture equipment corresponding to the node with the activation value exceeding the preset activation threshold is abnormal, thereby realizing reliability evaluation of mariculture equipment operation. The application constructs a dependency graph model based on a large model, regards the mariculture equipment in the marine ranching area as nodes that are related to each other, inputs a simulated fault signal, and infers a fault propagation probability and a chain reaction path using a Bayesian network, which directly solves the defect of “ignoring the dynamic mutual dependence relationship between the equipment” in the prior art. By monitoring the activation value of the node in the chain reaction path in real time and dynamically calculating the overall reliability level of all mariculture equipment, the method can capture and quantify the potential propagation effect of the fault in the equipment network in advance, significantly reduces the hysteresis of reliability evaluation, and makes the evaluation result closer to the dynamic reality of the complex marine ranching environment. In addition, by actively inputting a simulated fault signal, a forward-looking fault scenario deduction is realized, which not only can find potential risk points (nodes with activation values exceeding the limit) in the current state, but also can systematically reveal the chain reaction path and its influence range that may be triggered when a fault occurs, thereby providing a strong decision basis for proactive preventive maintenance and risk control. By comparing and analyzing the real-time calculated reliability level with the historical operation log data, when a similar fault scenario is matched, it can not only be determined that the overall operation state of all mariculture equipment is abnormal, but also accurately locate the specific source equipment (the equipment corresponding to the node with the activation value exceeding the limit) that causes the abnormality. This intelligent diagnosis based on historical experience greatly improves the accuracy and reliability of fault identification and state determination, and reduces false positives and false negatives.

[0177] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are merely examples of the present application and are not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A reliability assessment method for marine ranching equipment based on a large model, characterized in that, The mariculture farm comprises multiple types of cultivation equipment; The reliability evaluation method comprises: Obtaining operation parameters of each cultivation equipment in the mariculture farm through a preset sensor network; and analyzing according to the operation parameters to obtain an interaction data set between devices; Based on the interaction data set between devices, the cultivation equipment is taken as a node to construct a dependency graph model; wherein a large model is used in the process of constructing the dependency graph model, and the parameter quantity of the large model is greater than a preset quantity threshold; Based on the dependency graph model, at least one cultivation equipment is injected with a simulated fault signal, and a Bayesian network is used to infer a fault propagation probability to obtain at least one chain reaction path; the chain reaction path comprises multiple nodes; Monitoring the chain reaction path; and when the activation value of at least one node in the chain reaction path exceeds a preset activation threshold, calculating the reliability level of all cultivation equipment according to the path length of the chain reaction path and the influence weight of the chain reaction path on the dependency graph model; Obtaining historical operation log data; and comparing and analyzing the reliability level with the reliability data of each fault record sample in the historical operation log data; According to the comparison and analysis result, when a similar fault scene is matched in the historical operation log data, it is determined that the operation state of the cultivation equipment in the mariculture farm as a whole is abnormal, and it is also determined that the operation state of the cultivation equipment corresponding to the node with the activation value exceeding the preset activation threshold is abnormal, thereby realizing reliability evaluation of the cultivation equipment operation. 2.The reliability evaluation method of a large model-based mariculture facility according to claim 1, wherein, The type of the cultivation equipment comprises an oxygen supply device; and based on the dependency graph model, at least one cultivation equipment is injected with a simulated fault signal, and a Bayesian network is used to infer a fault propagation probability to obtain at least one chain reaction path, which comprises: A depth-first algorithm is used to perform segmentation processing on the dependency graph model based on the node corresponding to the oxygen supply device to obtain a subgraph associated with the oxygen supply device, thereby obtaining a subgraph node set; A dependency relationship matrix between nodes is constructed according to the nodes in the subgraph node set; the dependency relationship matrix is represented by an adjacency matrix; A simulated fault signal is injected into the node corresponding to the oxygen supply device to obtain a fault initial state; Based on the dependency relationship matrix and the fault initial state, a Bayesian network is used to infer a fault propagation probability distribution; According to the fault propagation probability distribution, the subgraph node set is traversed to obtain at least one chain reaction path. 3.The reliability evaluation method of a large model-based mariculture facility according to claim 1, wherein, The historical operation log data is obtained; The comparison and analysis of the reliability level with the reliability data of each fault record sample in the historical operation log data comprises: The historical operation log data is obtained; A plurality of fault record samples are extracted from the historical operation log data; wherein the sample information of each fault record sample comprises a fault starting time, a device ID, a fault type, and a duration; A fault tree analysis algorithm is used to calculate the fault probability of a sample device in the mariculture farm according to the sample information; Based on the fault probability of the sample device, a key device is determined; The reliability data of the mariculture farm as a whole is calculated through the key device; and According to the reliability data, each fault record sample is clustered to obtain a plurality of fault scenarios; The reliability level is compared and analyzed with the reliability data of each fault record sample in the historical operation log data; According to the comparison and analysis result, the fault scenario with the highest matching degree and the matching degree greater than the preset matching threshold is determined as the similar fault scenario. 4.The reliability evaluation method of a large model-based mariculture facility according to claim 1, wherein, The running parameters of each aquaculture device in the marine ranching are obtained through a preset sensor network, including: The water body dissolved oxygen content of the area where each aquaculture device in the marine ranching is located is obtained through a dissolved oxygen sensor to obtain oxygen content data; The motor load current of each aquaculture device in the marine ranching is obtained through a current sensor to obtain device load data; The water temperature data of the area where each aquaculture device in the marine ranching is located is obtained through a temperature sensor; The running parameters are obtained according to the water temperature data, oxygen content data and device load data.

5. The large model-based mariculture farm equipment reliability evaluation method of claim 4, wherein, The inter-device interaction data set is obtained by analyzing the running parameters, including: A data packet is generated based on the running parameters through a preset transmission protocol, and the data packet is transmitted to an edge computing node; wherein the format of the data packet is JSON format, and the data packet further contains the device ID, timestamp and parameter value corresponding to the running parameter; The edge computing node is controlled to use the MQTT protocol to aggregate the data packet to the cloud database, and the data packet is processed as time series data in the cloud database; The data packet is associated and analyzed through a preset association analysis model to obtain an association analysis result; and the Pearson correlation coefficient between the water temperature data, oxygen content data and device load data is calculated based on the data packet; The inter-device interaction data set is obtained by analyzing the association analysis result and the Pearson correlation coefficient.

6. The large model-based mariculture farm equipment reliability evaluation method of claim 5, wherein, Before the association analysis of the data packet through the preset association analysis model is performed to obtain the association analysis result, the data packet is further verified through a CRC32 verification algorithm; when the verification is passed, the verified data packet is obtained; when the verification fails, the edge computing node is controlled to use the MQTT protocol to aggregate the data packet to the cloud database again. 7.A reliability evaluation system for a large model-based mariculture facility, characterized by, The marine ranching contains multiple types of aquaculture devices; The reliability evaluation system includes an analysis module, a model construction module, an inference module, a calculation module, a comparison module and an evaluation module; The analysis module is used to obtain the running parameters of each aquaculture device in the marine ranching through a preset sensor network; and the inter-device interaction data set is obtained by analyzing the running parameters; The model construction module is used to construct a dependency graph model based on the inter-device interaction data set, taking the aquaculture device as a node; wherein a large model is used in the dependency graph model construction process, and the parameter quantity of the large model is greater than a preset quantity threshold; The inference module is used to input a simulated fault signal into at least one aquaculture device based on the dependency graph model, infer the fault propagation probability using a Bayesian network, and obtain at least one chain reaction path; the chain reaction path contains a plurality of nodes. The computing module is configured to monitor the chain reaction path, and when an activation value of at least one node in the chain reaction path exceeds a preset activation threshold, calculate a reliability level of the whole aquaculture equipment according to a path length of the chain reaction path and an influence weight of the chain reaction path on the dependency graph model; The comparison module is configured to obtain historical operation log data, and compare and analyze the reliability level with reliability data of each fault record sample in the historical operation log data; The evaluation module is configured to, according to a comparison and analysis result, when a similar fault scenario is matched in the historical operation log data, determine that an operation state of the whole aquaculture equipment in the marine ranching is abnormal, and simultaneously determine that an operation state of an aquaculture equipment corresponding to a node whose activation value exceeds the preset activation threshold is abnormal, so as to realize reliability evaluation of aquaculture equipment operation.

8. The large model-based mariculture farm equipment reliability assessment system of claim 7, wherein, The type of the aquaculture equipment includes an oxygen supply equipment; the inference module injects a simulated fault signal into at least one aquaculture equipment based on the dependency graph model, infers a fault propagation probability by using a Bayesian network, and obtains at least one chain reaction path, including: The inference module performs segmentation processing on the dependency graph model based on the node corresponding to the oxygen supply equipment by using a depth-first algorithm, obtains a subgraph associated with the oxygen supply equipment, and obtains a node set of the subgraph; A dependency relationship matrix between nodes is constructed according to the nodes in the node set of the subgraph; the dependency relationship matrix is represented by an adjacency matrix; A simulated fault signal is injected into the node corresponding to the oxygen supply equipment, and an initial fault state is obtained; A fault propagation probability distribution is inferred by using a Bayesian network based on the dependency relationship matrix and the initial fault state; At least one chain reaction path is obtained by traversing the node set of the subgraph according to the fault propagation probability distribution.

9. The large model-based mariculture farming facility reliability assessment system of claim 7, wherein, The comparison module obtains historical operation log data; The comparison module compares and analyzes the reliability level with reliability data of each fault record sample in the historical operation log data, including: The comparison module obtains historical operation log data; A plurality of fault record samples are extracted from the historical operation log data; sample information of each fault record sample includes a fault starting time, an equipment ID, a fault type, and a duration; A fault probability of a sample equipment in the marine ranching is calculated according to the sample information by using a fault tree analysis algorithm; A key equipment is determined based on the fault probability of the sample equipment; Reliability data of the whole marine ranching is calculated through the key equipment; According to the reliability data, each fault record sample is clustered to obtain a plurality of fault scenarios; The comparison module compares and analyzes the reliability level with reliability data of each fault record sample in the historical operation log data; According to a comparison and analysis result, a fault scenario with a highest matching degree and a matching degree greater than a preset matching threshold is determined as a similar fault scenario.

10. The large model-based mariculture farm equipment reliability evaluation system of claim 7, wherein, The analysis module obtains operation parameters of each aquaculture equipment in the marine ranching through a preset sensor network, including: The analysis module obtains oxygen content data of water in a region where each aquaculture equipment is located by using a dissolved oxygen sensor. The motor load current of each culture equipment in the mariculture farm is acquired through the current sensor, and equipment load data is obtained. The water temperature data of the area where each culture equipment in the mariculture farm is located is acquired through the temperature sensor. The operation parameters are obtained according to the water temperature data, the oxygen content data and the equipment load data.

Images