Submersible real-time monitoring method and device based on data analysis

Abstract

The invention discloses a submersible real-time monitoring method and device based on data analysis, and the method can capture the subtle change trend of a parameter before the parameter deviates from a normal threshold value through the continuous analysis of a real-time data flow, possibly recognizes that the energy of a certain specific frequency band of a vibration signal is slowly and continuously increased, and achieves the real-time monitoring of the submersible. Even if the absolute value does not reach an alarm threshold value, abnormal early warning can be carried out at a high probability, and the predictive diagnosis based on data mode evolution wins precious advance for operators, so that the operators can take preventive measures such as speed reduction, inspection and spare part preparation before the faults completely break out, and therefore, catastrophic accidents are avoided, and the safety of the operators is improved. And the task success rate and the safety of the submersible are greatly improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of submersible monitoring, and particularly relates to a submersible real-time monitoring method and device based on data analysis. BACKGROUND

[0002] Submersibles (including manned submersibles and unmanned submersibles) play a crucial role in deep-sea exploration, resource exploration, underwater engineering, military reconnaissance and other fields. Due to their extreme working environment (high pressure, low temperature, high corrosiveness), and once a failure occurs, rescue and maintenance are extremely difficult, and even may lead to submersible damage and personnel casualties, therefore, real-time and accurate fault monitoring and early warning of the submersible is the core link to ensure its safe operation.

[0003] The traditional submersible monitoring method mainly relies on threshold judgment and simple logic rules. For example, when the reading of a certain sensor (such as temperature, pressure) exceeds the preset safety threshold, the system triggers an alarm. Although this method is simple to implement, it has the following significant defects: Low fault diagnosis accuracy and high false positive and false negative rates: the systems of the submersible are interrelated and interdependent, and the abnormality of a single parameter may not directly represent a specific fault, while the small, coordinated changes of multiple parameters may be a precursor to a serious fault. The single parameter judgment method based on threshold cannot capture this complex, multi-dimensional correlation, resulting in insufficient ability to identify complex fault patterns, and is prone to false positives (misjudging normal fluctuations as faults) and false negatives (failing to identify real faults).

[0004] Lack of predictability and real-time lag: the traditional method is mostly "post-alarm", that is, it only issues an alarm after the fault has occurred or the parameter has deviated significantly from the normal range. It cannot predict the occurrence of a fault by analyzing the subtle changes in data trends in the early stages of the fault, thus missing the best opportunity for intervention and treatment. SUMMARY

[0005] The present application provides a submersible real-time monitoring method and device based on data analysis to solve the problems of low fault diagnosis accuracy, high false positive and false negative rates, lack of predictability and real-time lag in the prior art.

[0006] The first aspect of the present application provides a submersible real-time monitoring method based on data analysis, comprising: collecting experimental operation data of the submersible in the laboratory and the real submersible operation state when the experimental operation data is collected; learning the data relationship between the experimental operation data and the real submersible operation state when the experimental operation data is collected using a fault diagnosis model, and obtaining a fault diagnosis model with fault identification capability; Actual operation data is collected during actual operation of the submersible, and the fault diagnosis model with fault recognition capability is dispatched to analyze the actual operation data, so as to obtain a submersible fault monitoring result; Real-time early warning information of the submersible is generated according to the submersible fault monitoring result, and the real-time early warning information of the submersible is visually displayed, so as to complete real-time monitoring of the submersible based on data analysis.

[0007] In a possible design, experimental operation data of a submersible in a laboratory and a real submersible operation state when the experimental operation data is collected are collected, including: The experimental operation data at N sampling time points are continuously collected based on a preset data sampling frequency; wherein the experimental operation data at least include submersible temperature, pressure, vibration, current and voltage; The real submersible operation state at the Nth sampling time point is collected, so as to obtain the real submersible operation state when the experimental operation data is collected; The experimental operation data and the real submersible operation state when the experimental operation data is collected are repeatedly collected.

[0008] In a possible design, a fault diagnosis model is used to learn the data relationship between the experimental operation data and the real submersible operation state when the experimental operation data is collected, so as to obtain a fault diagnosis model with fault recognition capability, including: A CNN-LSTM model is used to construct a fault diagnosis model, and model parameters of the fault diagnosis model are vector encoded, so as to obtain a model parameter vector; According to the experimental operation data and the real submersible operation state when the experimental operation data is collected, a loss function value corresponding to the model parameter vector is obtained; The loss function value corresponding to the model parameter vector is converted into fitness, so as to obtain a fitness corresponding to each model parameter vector, and a model parameter vector with the maximum fitness is determined as a first target optimal vector; A self-adaptive adaptive sine-cosine interaction strategy is used to perform sine-cosine local exploration on the model parameter vector, so as to obtain a model parameter vector after sine-cosine local exploration; A multi-point social learning strategy is used to perform social information learning on the model parameter vector after the sine-cosine local exploration, so as to obtain a model parameter vector after social information learning; Based on the first target optimal vector, a local fluctuation exploration strategy is used to perform local fluctuation exploration on the model parameter vector after the social information learning, so as to obtain a model parameter vector after local fluctuation exploration; A single-point diffusion exploration strategy is used to perform a global search on the model parameter vector after the local fluctuation exploration, so as to obtain the model parameter vector after the global search. Obtain the current training state; wherein, the current training state includes a training completed state or a training incomplete state; When the current training state is the training completed state, the second target optimal vector is determined based on the model parameter vector after global search, and a fault diagnosis model with fault identification capability is obtained based on the second target optimal vector. If the current training state is incomplete, then return to the step of obtaining the optimal vector for the first target.

[0009] In one possible design approach, an adaptive sine / cosine interaction strategy is used to perform sine / cosine local exploration on the model parameter vector, resulting in the model parameter vector after sine / cosine local exploration:

[0010]

[0011] in, Indicates the first t During the training process, the first i A vector of model parameters, Indicates the first i The model parameter vector after local exploration of sine and cosine curves. i =1,2,…,NP, where NP represents the total number of model parameter vectors. Represents the adaptive fit coefficient. Represents a random number between (0,1). Represents (0,2) π A random number between ) π Represents pi (π). Represents a random number between (0,1). Represented as Other model parameter vectors matched randomly, Represents a random number between (0,1). Represents (0,2) π A random number between ) Represents a random number between (0,1). express Adaptability, express The degree of adaptability.

[0012] In one possible design approach, a multi-point social learning strategy is employed to learn social information from the model parameter vectors obtained after the local exploration of sine and cosine parameters, resulting in the model parameter vectors after social information learning:

[0013]

[0014]

[0015] in, Indicates the first t During the training process, the first n The model parameter vector after local exploration of sine and cosine curves. Indicates the first n The model parameter vector after learning from social information. This represents the first social learning factor. This represents the second social learning factor. This represents the model parameter vector after the first random sine and cosine local exploration. This represents the model parameter vector after the second random sine and cosine local exploration. Represents a random number between (0,1). Represents a random number between (0,1). express Adaptability, express Adaptability, express The degree of adaptability.

[0016] In one possible design approach, based on the first target optimal vector, a local fluctuation exploration strategy is used to perform local fluctuation exploration on the model parameter vector after learning the social information, resulting in the model parameter vector after local fluctuation exploration:

[0017]

[0018]

[0019]

[0020] in, Indicates the first t During the training process, the first m The model parameter vector after learning from social information. Indicates the first m The model parameter vector after exploring local fluctuations. Indicates the first t +1 training session m The exploration speed of the model parameter vector after learning from social information. Indicates the firstt During the training process, the first m The exploration speed of the model parameter vector after learning from social information. Indicates the first information exploration factor. This represents the second information exploration factor. Represents a random number between (0,1). Represents a random number between (0,1). This represents the optimal vector for the first objective. Indicates the first t -1 training session m The model parameter vector after learning from social information. Indicates the first volatility factor. Indicates the second volatility factor. express The historical optimal state.

[0021] In one possible design approach, a single-point diffusion exploration strategy is used to perform a global search on the model parameter vector after the local fluctuation exploration, resulting in the following model parameter vector after the global search: The diffusion exploration range control factor is:

[0022] in, Indicates the first k The diffusion exploration range control factor corresponding to the model parameter vector after local fluctuation exploration. Indicates the basic diffusion search range parameter. Indicates the first k The fitness of the model parameter vector after exploring local fluctuations. NP represents the minimum fitness of the model parameter vector after exploring all local fluctuations, and NP represents the total number of model parameter vectors. Based on the diffusion exploration range control factor, a global search is performed on the model parameter vector after the local fluctuation exploration to obtain the model parameter vector after the global search.

[0023] in, Indicates the first t During the training process, the first k The model parameter vector after exploring the local fluctuations is the first... d 3D model parameters, Indicates the first k The model parameter vector after the nth global search d 3D model parameters, d =1,2,…,L, where L represents the total number of model parameters in the model parameter vector. express and A random number between [a certain number of points].

[0024] In one possible design approach, actual operational data is collected during the submersible's actual operation, and a fault diagnosis model with fault identification capabilities is scheduled to analyze the actual operational data to obtain submersible fault monitoring results, including: During the actual operation of the submersible, actual operating data is collected, and the fault diagnosis model with fault identification capability is scheduled to perform data analysis on the actual operating data to obtain the probability distribution output by the fault diagnosis model with fault identification capability. Based on the probability distribution, the category with the highest probability is determined as the submersible fault monitoring result.

[0025] In one possible design approach, real-time early warning information for the submersible is generated based on the submersible fault monitoring results, and the real-time early warning information is visualized, including: When the submersible's fault monitoring result is classified as an abnormal operation, a real-time warning message for the submersible is generated based on the actual operating data and the submersible's fault monitoring result. This real-time warning message is then displayed on the submersible's built-in screen or transmitted to a device designated by the staff via the submersible's wireless communication function.

[0026] A second aspect of the present invention provides a real-time monitoring device for submersibles based on data analysis, comprising: The experimental data acquisition module is used to collect experimental operation data of the submersible in the laboratory and the actual operating status of the submersible when the experimental operation data is collected; The data relationship learning module is used to learn the data relationship between the experimental operation data and the actual submersible operation status when the experimental operation data is collected using a fault diagnosis model, so as to obtain a fault diagnosis model with fault identification capability. The data analysis module is used to collect actual operating data during the actual operation of the submersible, and to schedule the fault diagnosis model with fault identification capability to perform data analysis on the actual operating data to obtain submersible fault monitoring results. The real-time monitoring and early warning module is used to generate real-time early warning information for the submersible based on the submersible fault monitoring results, and to visualize the real-time early warning information, thereby completing real-time monitoring of the submersible based on data analysis.

[0027] This invention provides a data analysis-based real-time monitoring method and device for submersibles. By continuously analyzing real-time data streams, it can capture subtle trends in parameter changes before they deviate from normal thresholds. It may identify a slow and continuous increase in energy at a specific frequency band of vibration signals, and even if the absolute value is far from the alarm threshold, it can provide an early warning of anomalies with a high probability. This predictive diagnosis based on data pattern evolution gives operators valuable lead time, enabling them to take preventative measures such as deceleration, inspection, and preparation of spare parts before a complete failure occurs, thereby avoiding catastrophic accidents and greatly improving the mission success rate and safety of submersibles. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0029] Figure 1 A flowchart illustrating a real-time monitoring method for submersibles based on data analysis, provided as an embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of a real-time monitoring device for a submersible based on data analysis, provided as an embodiment of the present invention.

[0031] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0032] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0033] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0034] like Figure 1 As shown, this embodiment of the invention provides a real-time monitoring method for submersibles based on data analysis, including: S101. Collect experimental operation data of the submersible in the laboratory and the actual operating status of the submersible when the experimental operation data is collected.

[0035] For example, in a laboratory environment, a series of simulated normal and fault conditions were tested on the target submersible. The data sampling frequency was set to 10Hz, meaning 10 data points were collected per second. Each test collected N=100 consecutive sampling time points (i.e., a 10-second data segment) of experimental data, including: propulsion motor temperature (°C), sealed chamber pressure (Pa), and pump vibration acceleration (m / s²). 2 The data includes the main circuit current (A) and control voltage (V). After each 10-second data acquisition segment, experts or a pre-set fault injection device record the actual operating status of the submersible during that time period, using labels such as "normal," "motor overheating," "pressure seal failure," and "hydraulic pump malfunction." This process is repeated thousands of times, covering various typical operating conditions and fault modes, to build a labeled dataset containing 100,000 samples.

[0036] It is worth noting that each type of data can be collected by sensors installed at various locations on the submersible that require monitoring. For example, when collecting current, the main circuit current or the current of each branch circuit can be collected. Those skilled in the art can configure the system according to actual needs.

[0037] S102. A fault diagnosis model is used to learn the data relationship between the experimental operation data and the actual submersible operation state when the experimental operation data is collected, so as to obtain a fault diagnosis model with fault identification capability.

[0038] During the learning process, the model not only learned the obvious differences between faults and normal conditions, but more importantly, it learned the gradual transition from normal to fault. Therefore, when equipment performance has just begun to degrade and parameters have not yet reached dangerous levels, the model can identify this abnormal state. This represents a leap from "post-event alarms" to "pre-event warnings." Operators have ample time to take preventative maintenance measures, such as adjusting operating conditions, preparing spare parts, or returning to base early, thereby avoiding sudden failures and greatly improving the submersible's mission success rate and safety.

[0039] S103. During the actual operation of the submersible, collect actual operating data and schedule the fault diagnosis model with fault identification capability to perform data analysis on the actual operating data to obtain submersible fault monitoring results.

[0040] For example, when a submersible is performing a mission, its onboard data acquisition system collects real-time data on temperature, pressure, vibration, current, and voltage at a frequency of 10Hz. The latest 100 data points (i.e., data from the last 10 seconds) are organized into a 100x5 matrix and input into a pre-trained fault diagnosis model in real time. After forward propagation, the model outputs a probability vector, for example [0.85, 0.08, 0.05, 0.02], corresponding to the probabilities of "normal," "motor overheating," "pressure seal failure," and "hydraulic pump malfunction," respectively. The category with the highest probability, "normal," is selected as the current submersible fault monitoring result.

[0041] S104. Generate real-time early warning information for the submersible based on the submersible fault monitoring results, and visualize the real-time early warning information for the submersible to complete the real-time monitoring of the submersible based on data analysis.

[0042] Suppose that at a certain moment, the model output probability becomes [0.15, 0.78, 0.05, 0.02]. At this time, the probability of "motor overheating" is the highest and exceeds the preset warning threshold (e.g., 0.7), which is determined to be an abnormal operation category. Then, based on the actual operating data and the submersible fault monitoring results, real-time warning information for the submersible can be generated and displayed on the submersible's built-in screen or transmitted to a device designated by the operator via the submersible's wireless communication function. During the real-time warning information generation process, actual operating data and submersible fault monitoring results can be filled in using a preset template.

[0043] This invention, through continuous analysis of real-time data streams, can capture subtle trends in parameter changes before they deviate from normal thresholds. For example, the model may identify a slow, continuous increase in energy within a specific frequency band of a vibration signal, even if its absolute value is far below the alarm threshold. In this case, the model will output a warning of "early bearing wear" with a high probability. This predictive diagnostic based on data pattern evolution provides operators with valuable lead time, enabling them to take preventative measures such as slowing down, inspecting, and preparing spare parts before a full-blown failure occurs, thereby avoiding catastrophic accidents and significantly improving the mission success rate and safety of the submersible.

[0044] Optionally, when a fault is detected, the system automatically generates a warning report containing structured information such as the fault type, probability of occurrence, key abnormal parameters, and recommended handling measures. For visualization, the system uses the submersible's built-in screen and remote monitoring equipment to present information intuitively through a graphical interface (such as highlighted instrument panels, marked abnormal points on graphs, and flashing fault locations in 3D models), making it easy for operators to understand at a glance. Simultaneously, utilizing wireless communication technology, warning information can be instantly pushed to the surface command center, engineering team, and even the mobile terminals of relevant personnel, ensuring the real-time and widespread dissemination of information, breaking down information silos, enabling collaborative decision-making among multiple parties, significantly shortening fault response time, and improving the efficiency and coordination of emergency response.

[0045] In one possible design approach, collecting experimental operation data of the submersible in the laboratory and the actual operating status of the submersible at the time of collecting the experimental operation data includes: Based on a preset data sampling frequency, experimental operation data is continuously collected at N sampling time points; wherein, the experimental operation data includes at least the submersible temperature, pressure, vibration, current and voltage; Collect the actual submersible operating status at the Nth sampling time point to obtain the actual submersible operating status when the experimental operation data is obtained; The experimental operation data and the actual operating status of the submersible during the data collection were collected repeatedly.

[0046] In one possible design approach, a fault diagnosis model is used to learn the data relationship between the experimental operation data and the actual submersible operating state when the experimental operation data was collected, thereby obtaining a fault diagnosis model with fault identification capabilities, including: A fault diagnosis model is constructed using a CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory Network) model, and the model parameters of the fault diagnosis model are vector-encoded to obtain a model parameter vector. For example, the model parameters can be randomly initialized between the upper and lower limits, and the initialized model parameters can be encoded into vectors to obtain a model parameter vector. Multiple model parameter vectors can be obtained repeatedly and used for training.

[0047] Traditional methods only perform threshold judgments on single parameters, ignoring the complex relationships between parameters. This invention employs a CNN-LSTM hybrid model as the core diagnostic engine. CNN (Convolutional Neural Network) can automatically extract local spatial features and patterns from raw, multi-dimensional sensor data (temperature, pressure, vibration, current, voltage, etc.), for example, the correlation between specific frequency components of vibration signals and motor bearing wear. LSTM (Long Short-Term Memory Network) excels at processing time-series data, capturing the dynamic patterns and long-term dependencies of these features over time; for example, the combined trend of a slow increase in temperature and a gradual decrease in pressure may indicate a progressive failure in the cooling system. Through this dual mechanism of "spatial feature extraction + time-series modeling," this invention can deeply mine the fault information contained in the data, accurately identify complex, multi-parameter coupled fault modes, greatly reduce the false alarm rate and false negative rate, and make the diagnostic results more reliable.

[0048] Based on the experimental data and the actual operating state of the submersible when the experimental data was collected, the loss function value corresponding to the model parameter vector is obtained; for example, the cross-entropy loss function or the root mean square loss function can be used to obtain the loss function value.

[0049] The loss function value corresponding to the model parameter vector is converted into fitness, and the fitness of each model parameter vector is obtained. The model parameter vector with the largest fitness is determined as the first target optimal vector. For example, the loss function value can be negative or its reciprocal to convert it into fitness.

[0050] An adaptive sine-cosine interaction strategy is used to perform sine-cosine local exploration on the model parameter vector to obtain the model parameter vector after sine-cosine local exploration.

[0051] A multi-point social learning strategy is used to learn social information from the model parameter vectors after the local exploration of sine and cosine, resulting in a model parameter vector after social information learning.

[0052] Based on the first target optimal vector, a local fluctuation exploration strategy is used to explore the model parameter vector after learning the social information, so as to obtain the model parameter vector after local fluctuation exploration.

[0053] A single-point diffusion exploration strategy is used to perform a global search on the model parameter vector after the local fluctuation exploration, resulting in a global search model parameter vector.

[0054] Obtain the current training status; wherein, the current training status includes a training completed status or a training incomplete status. For example, if the current number of training iterations is greater than or equal to the preset maximum number of training iterations, it is determined to be a training completed status; otherwise, it is determined to be a training incomplete status.

[0055] When the current training state is training complete, the second target optimal vector is determined based on the model parameter vector obtained after global search, and a fault diagnosis model with fault identification capability is obtained based on the second target optimal vector. For example, the model parameters in the second target optimal vector can be used as the final parameters of the fault diagnosis model to obtain a fault diagnosis model with fault identification capability.

[0056] If the current training state is incomplete, then return to the step of obtaining the optimal vector for the first target.

[0057] Optionally, after a vector changes, out-of-bounds handling can be performed to ensure parameter validity.

[0058] CNN-LSTM models have numerous parameters, making them prone to getting stuck in local optima during training, leading to poor model performance. This invention provides a hybrid optimization strategy comprising four complementary stages, effectively solving this problem: By organically combining four strategies, a closed-loop optimization process of "wide-area exploration → information exchange → local refinement → global transition" is formed. This not only converges faster than traditional optimization algorithms (such as gradient descent and standard genetic algorithms) and reduces training time by more than 30%, but also yields superior model parameters. This significantly improves key performance indicators such as accuracy, recall, and F1 score of the fault diagnosis model, resulting in stronger generalization ability and better adaptability to various unknown and complex operating conditions in real-world scenarios.

[0059] In one possible design approach, an adaptive sine / cosine interaction strategy is used to perform sine / cosine local exploration on the model parameter vector, resulting in the model parameter vector after sine / cosine local exploration:

[0060]

[0061] in, Indicates the first t During the training process, the first i A vector of model parameters, Indicates the first i The model parameter vector after local exploration of sine and cosine curves. i =1,2,…,NP, where NP represents the total number of model parameter vectors. Represents the adaptive fit coefficient. Represents a random number between (0,1). Represents (0,2) π A random number between ) π Represents pi (π). Represents a random number between (0,1). Represented as Other model parameter vectors matched randomly, Represents a random number between (0,1). Represents (0,2) π A random number between ) Represents a random number between (0,1). express Adaptability, express The degree of adaptability.

[0062] In the initial optimization phase, this strategy utilizes the periodic fluctuations of sine and cosine functions for extensive local exploration, which helps the population quickly disperse to different regions of the solution space, enhancing the breadth of the global search. Its adaptive coefficients can dynamically adjust the exploration step size according to the iteration process, balancing the breadth of exploration in the early stage with the accuracy development in the later stage.

[0063] In one possible design approach, a multi-point social learning strategy is employed to learn social information from the model parameter vectors obtained after the local exploration of sine and cosine parameters, resulting in the model parameter vectors after social information learning:

[0064]

[0065]

[0066] in, Indicates the first t During the training process, the first n The model parameter vector after local exploration of sine and cosine curves. Indicates the first n The model parameter vector after learning from social information. This represents the first social learning factor. This represents the second social learning factor. This represents the model parameter vector after the first random sine and cosine local exploration. This represents the model parameter vector after the second random sine and cosine local exploration. Represents a random number between (0,1). Represents a random number between (0,1). express Adaptability, express Adaptability, express The degree of adaptability.

[0067] This stage simulates a social learning mechanism, where each solution (model parameter vector) learns not only from the current best solution but also from other high-performing random individuals. This multi-source communication mechanism accelerates the spread of high-quality genes (effective parameter combinations) in the population, preventing the algorithm from converging prematurely to suboptimal solutions and improving both convergence speed and solution quality.

[0068] In one possible design approach, based on the first target optimal vector, a local fluctuation exploration strategy is used to perform local fluctuation exploration on the model parameter vector after learning the social information, resulting in the model parameter vector after local fluctuation exploration:

[0069]

[0070]

[0071]

[0072] in, Indicates the first t During the training process, the first m The model parameter vector after learning from social information. Indicates the first m The model parameter vector after exploring local fluctuations. Indicates the first t +1 training session m The exploration speed of the model parameter vector after learning from social information. Indicates the first t During the training process, the first m The exploration speed of the model parameter vector after learning from social information. Indicates the first information exploration factor. This represents the second information exploration factor. Represents a random number between (0,1). Represents a random number between (0,1). This represents the optimal vector for the first objective. Indicates the first t -1 training session m The model parameter vector after learning from social information. Indicates the first volatility factor. Indicates the second volatility factor. express The historical optimal state.

[0073] This strategy introduces the concepts of velocity and historical best states, enabling a refined, inertial, and memory-based fluctuation search for solutions near the current best. This is analogous to conducting a thorough search around the most promising location, effectively uncovering better solutions near local optima and significantly improving the algorithm's local optimization accuracy.

[0074] In one possible design approach, a single-point diffusion exploration strategy is used to perform a global search on the model parameter vector after the local fluctuation exploration, resulting in the following model parameter vector after the global search: The diffusion exploration range control factor is:

[0075] in, Indicates the first k The diffusion exploration range control factor corresponding to the model parameter vector after local fluctuation exploration. Indicates the basic diffusion search range parameter. Indicates the first k The fitness of the model parameter vector after exploring local fluctuations. NP represents the minimum fitness of the model parameter vector after exploring all local fluctuations, and NP represents the total number of model parameter vectors. Based on the diffusion exploration range control factor, a global search is performed on the model parameter vector after the local fluctuation exploration to obtain the model parameter vector after the global search.

[0076] in, Indicates the first t During the training process, the first k The model parameter vector after exploring the local fluctuations is the first... d 3D model parameters, Indicates the first k The model parameter vector after the nth global search d 3D model parameters, d =1,2,…,L, where L represents the total number of model parameters in the model parameter vector. express and A random number between [a certain number of points].

[0077] To prevent the algorithm from getting trapped in local optima, this strategy dynamically adjusts the diffusion search range of each solution in the later stages of each iteration based on its fitness (quality). Poorly performing solutions receive a larger diffusion range, giving them the opportunity to escape the current region and explore entirely new unknown spaces; well-performing solutions are fine-tuned within a smaller range. This adaptive global search mechanism effectively maintains the diversity of the population and is key to avoiding premature convergence and ensuring the discovery of the globally optimal solution. Simulated annealing can also be used during the global exploration process to control the algorithm and ensure its training speed.

[0078] In one possible design approach, actual operational data is collected during the submersible's actual operation, and a fault diagnosis model with fault identification capabilities is scheduled to analyze the actual operational data to obtain submersible fault monitoring results, including: During the actual operation of the submersible, actual operating data is collected, and the fault diagnosis model with fault identification capability is scheduled to perform data analysis on the actual operating data to obtain the probability distribution output by the fault diagnosis model with fault identification capability.

[0079] Based on the probability distribution, the category with the highest probability is determined as the submersible fault monitoring result.

[0080] In one possible design approach, real-time early warning information for the submersible is generated based on the submersible fault monitoring results, and the real-time early warning information is visualized, including: When the submersible's fault monitoring result is classified as an abnormal operation, a real-time warning message for the submersible is generated based on the actual operating data and the submersible's fault monitoring result. This real-time warning message is then displayed on the submersible's built-in screen or transmitted to a device designated by the staff via the submersible's wireless communication function.

[0081] like Figure 2 As shown, based on the same inventive concept, this embodiment of the invention provides a real-time monitoring device for submersibles based on data analysis, comprising: The experimental data acquisition module 201 is used to collect experimental operation data of the submersible in the laboratory and the actual operating status of the submersible when the experimental operation data is collected. The data relationship learning module 202 is used to learn the data relationship between the experimental operation data and the actual submersible operation state when the experimental operation data is collected using a fault diagnosis model, so as to obtain a fault diagnosis model with fault identification capability. The data analysis module 203 is used to collect actual operating data during the actual operation of the submersible, and to schedule the fault diagnosis model with fault identification capability to perform data analysis on the actual operating data to obtain submersible fault monitoring results. The real-time monitoring and early warning module 204 is used to generate real-time early warning information for the submersible based on the submersible fault monitoring results, and to visualize the real-time early warning information for the submersible, thereby completing real-time monitoring of the submersible based on data analysis.

[0082] The real-time monitoring device for submersibles based on data analysis provided in this embodiment of the invention can execute the above-described method and technical solution. Its principle and beneficial technical effects are similar, and will not be repeated here.

[0083] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that the invention is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A real-time monitoring method for submersibles based on data analysis, characterized in that, include: Collect experimental operation data of the submersible in the laboratory and the actual operating status of the submersible at the time of collecting the experimental operation data; A fault diagnosis model is used to learn the data relationship between the experimental operation data and the actual submersible operation state when the experimental operation data is collected, so as to obtain a fault diagnosis model with fault identification capability. During the actual operation of the submersible, actual operating data is collected, and the fault diagnosis model with fault identification capability is scheduled to perform data analysis on the actual operating data to obtain submersible fault monitoring results. Based on the submersible fault monitoring results, real-time early warning information for the submersible is generated and visualized, thus completing real-time monitoring of the submersible based on data analysis. A fault diagnosis model is used to learn the data relationship between the experimental operation data and the actual submersible operating state when the experimental operation data was collected, thereby obtaining a fault diagnosis model with fault identification capabilities, including: A fault diagnosis model is constructed using a CNN-LSTM model, and the model parameters of the fault diagnosis model are vector-encoded to obtain the model parameter vector. Based on the experimental data and the actual submersible operating state when the experimental data was collected, the loss function value corresponding to the model parameter vector is obtained; The loss function values ​​corresponding to the model parameter vectors are converted into fitness values ​​to obtain the fitness values ​​for each model parameter vector, and the model parameter vector with the largest fitness value is determined as the first target optimal vector. An adaptive sine-cosine interaction strategy is used to perform sine-cosine local exploration on the model parameter vector to obtain the model parameter vector after sine-cosine local exploration. A multi-point social learning strategy is used to learn social information from the model parameter vectors after the local exploration of sine and cosine, resulting in a model parameter vector after social information learning. Based on the first target optimal vector, a local fluctuation exploration strategy is used to explore the model parameter vector after learning the social information, so as to obtain the model parameter vector after local fluctuation exploration. A single-point diffusion exploration strategy is used to perform a global search on the model parameter vector after the local fluctuation exploration, so as to obtain the model parameter vector after the global search. Obtain the current training state; wherein, the current training state includes a training completed state or a training incomplete state; When the current training state is the training completed state, the second target optimal vector is determined based on the model parameter vector after global search, and a fault diagnosis model with fault identification capability is obtained based on the second target optimal vector. If the current training state is incomplete, then return to the step of obtaining the optimal vector for the first target.

2. The real-time monitoring method for submersibles based on data analysis according to claim 1, characterized in that, Collect experimental operation data of the submersible in the laboratory and the actual operating status of the submersible at the time of data collection, including: Based on a preset data sampling frequency, experimental operation data is continuously collected at N sampling time points; wherein, the experimental operation data includes at least the submersible temperature, pressure, vibration, current and voltage; Collect the actual submersible operating status at the Nth sampling time point to obtain the actual submersible operating status when the experimental operation data is obtained; The experimental operation data and the actual operating status of the submersible during the data collection were collected repeatedly.

3. The real-time monitoring method for submersibles based on data analysis according to claim 1, characterized in that, An adaptive sine-cosine interaction strategy is used to perform sine-cosine local exploration on the model parameter vector, resulting in the following model parameter vector after sine-cosine local exploration: in, Indicates the first t During the training process, the first i A vector of model parameters, Indicates the first i The model parameter vector after local exploration of sine and cosine curves. i =1,2,…,NP, where NP represents the total number of model parameter vectors. Represents the adaptive fit coefficient. Represents a random number between (0,1). Represents (0,2) π A random number between ) π Represents pi (π). Represents a random number between (0,1). Represented as Other model parameter vectors matched randomly, Represents a random number between (0,1). Represents (0,2) π A random number between ) Represents a random number between (0,1). express Adaptability, express The degree of adaptability.

4. The real-time monitoring method for submersibles based on data analysis according to claim 3, characterized in that, A multi-point social learning strategy is used to learn social information from the model parameter vectors obtained after the local exploration of sine and cosine parameters, resulting in the model parameter vectors after social information learning: in, Indicates the first t During the training process, the first n The model parameter vector after local exploration of sine and cosine curves. Indicates the first n The model parameter vector after learning from social information. This represents the first social learning factor. This represents the second social learning factor. This represents the model parameter vector after the first random sine and cosine local exploration. This represents the model parameter vector after the second random sine and cosine local exploration. Represents a random number between (0,1). Represents a random number between (0,1). express Adaptability, express Adaptability, express The degree of adaptability.

5. The real-time monitoring method for submersibles based on data analysis according to claim 4, characterized in that, Based on the first target optimal vector, a local fluctuation exploration strategy is used to explore the local fluctuations of the model parameter vector after learning the social information, resulting in the following model parameter vector after local fluctuation exploration: in, Indicates the first t During the training process, the first m The model parameter vector after learning from social information. Indicates the first m The model parameter vector after exploring local fluctuations. Indicates the first t +1 training session m The exploration speed of the model parameter vector after learning from social information. Indicates the first t During the training process, the first m The exploration speed of the model parameter vector after learning from social information. Indicates the first information exploration factor. This represents the second information exploration factor. Represents a random number between (0,1). Represents a random number between (0,1). This represents the optimal vector for the first objective. Indicates the first t -1 training session m The model parameter vector after learning from social information. Indicates the first volatility factor. Indicates the second volatility factor. express The historical optimal state.

6. The real-time monitoring method for submersibles based on data analysis according to claim 5, characterized in that, A single-point diffusion exploration strategy is used to perform a global search on the model parameter vector after the local fluctuation exploration, resulting in the following model parameter vector after the global search: The diffusion exploration range control factor is: in, Indicates the first k The diffusion exploration range control factor corresponding to the model parameter vector after local fluctuation exploration. Indicates the basic diffusion search range parameter. Indicates the first k The fitness of the model parameter vector after exploring local fluctuations. NP represents the minimum fitness of the model parameter vector after exploring all local fluctuations, and NP represents the total number of model parameter vectors. Based on the diffusion exploration range control factor, a global search is performed on the model parameter vector after the local fluctuation exploration to obtain the model parameter vector after the global search. in, Indicates the first t During the training process, the first k The model parameter vector after exploring the local fluctuations is the first... d 3D model parameters, Indicates the first k The model parameter vector after the global search is the first... d 3D model parameters, d =1,2,…,L, where L represents the total number of model parameters in the model parameter vector. express and A random number between [a certain number of points].

7. The real-time monitoring method for submersibles based on data analysis according to claim 1, characterized in that, During the actual operation of the submersible, actual operational data is collected, and the fault diagnosis model with fault identification capabilities is used to analyze the actual operational data to obtain submersible fault monitoring results, including: During the actual operation of the submersible, actual operating data is collected, and the fault diagnosis model with fault identification capability is scheduled to perform data analysis on the actual operating data to obtain the probability distribution output by the fault diagnosis model with fault identification capability. Based on the probability distribution, the category with the highest probability is determined as the submersible fault monitoring result.

8. The real-time monitoring method for submersibles based on data analysis according to claim 1, characterized in that, Based on the submersible fault monitoring results, real-time early warning information for the submersible is generated and visualized, including: When the submersible's fault monitoring result is classified as an abnormal operation, a real-time warning message for the submersible is generated based on the actual operating data and the submersible's fault monitoring result. This real-time warning message is then displayed on the submersible's built-in screen or transmitted to a device designated by the staff via the submersible's wireless communication function.

9. A data analysis-based real-time monitoring device for submersibles, capable of executing the data analysis-based real-time monitoring method for submersibles as described in any one of claims 1 to 8, characterized in that, include: The experimental data acquisition module is used to collect experimental operation data of the submersible in the laboratory and the actual operating status of the submersible when the experimental operation data is collected; The data relationship learning module is used to learn the data relationship between the experimental operation data and the actual submersible operation status when the experimental operation data is collected using a fault diagnosis model, so as to obtain a fault diagnosis model with fault identification capability. The data analysis module is used to collect actual operating data during the actual operation of the submersible, and to schedule the fault diagnosis model with fault identification capability to perform data analysis on the actual operating data to obtain submersible fault monitoring results. The real-time monitoring and early warning module is used to generate real-time early warning information for the submersible based on the submersible fault monitoring results, and to visualize the real-time early warning information, thereby completing real-time monitoring of the submersible based on data analysis.

Images