A method for automatic operation monitoring of a high altitude platform process system

Abstract

This invention discloses an automated operation monitoring method for an aerial work platform process system, belonging to the field of automated data monitoring technology. The method includes: collecting historical multi-source heterogeneous data and its actual operating status; using a machine learning model to learn data relationships and obtain a system operation monitoring model; collecting real-time multi-source heterogeneous data and inputting it into the model to obtain output; and generating an automated operation monitoring report. The model training employs a hybrid optimization algorithm: based on interactive information, a fractional-order exploration strategy, a social learning exploration strategy, and a spatial tumbling exploration strategy are sequentially used for optimization, iterating until a termination condition is met. The parameters in the final target parameter combination are used as the final parameters of the machine learning model. This invention, by integrating optimization algorithms with multiple exploration strategies, can adaptively search for the optimal model configuration, improve model accuracy and generalization ability, and achieve precise monitoring and early warning of the operating status of the aerial work platform process system.

Description

Technical Field

[0001] This invention relates to the field of automated data monitoring technology, specifically to an automated operation monitoring method for a high-altitude platform process system. Background Technology

[0002] A high-altitude test facility is a crucial large-scale ground-based testing facility that simulates the high-altitude flight environment (pressure, temperature, Mach number, etc.) of an aircraft to conduct performance, functional, and reliability tests on aero engines. Its process system is an extremely complex and vast system, typically including multiple subsystems such as fuel and gas supply systems, heating or cooling systems, power output loading systems, measurement and control systems, and data acquisition and analysis systems. These systems are interconnected and closely related; an anomaly in any link can lead to the interruption of the entire test, or even cause costly engine or equipment damage and safety accidents.

[0003] Currently, the operation monitoring of high-altitude test station process systems mainly relies on threshold alarm methods, which set fixed upper and lower limits for key process parameters (such as pressure, temperature, flow rate, and rotational speed). When a parameter exceeds the threshold, the system triggers an audible and visual alarm. This method is simple and direct, but it only alarms after a fault has occurred or when the parameter has significantly deviated from the normal range. It cannot provide early warning, nor can it detect situations where parameters are within the threshold but have already shown abnormal trends or pattern changes, such as slow parameter drift or periodic oscillations, which are often precursors to potential faults. In addition, traditional methods rely on human experience for fault diagnosis, resulting in low monitoring efficiency, high subjectivity, and difficulty in adapting to the dynamic changes of multiple coupled parameters in complex systems. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to overcome the defects of the prior art and provide an automated operation monitoring method for a high-altitude platform process system, which aims to solve the problems of delayed early warning and low monitoring efficiency in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An automated operation monitoring method for a high-altitude platform process system includes the following steps: Collect multiple historical multi-source heterogeneous data corresponding to the high-altitude test platform process system and the actual operating status corresponding to the historical multi-source heterogeneous data; wherein, the actual operating status includes normal operating status and various fault operating status; A machine learning model is used to learn the data relationship between the historical multi-source heterogeneous data and the actual operating status to obtain a system operation monitoring model; Collect real-time multi-source heterogeneous data corresponding to the high-altitude platform process system, and use the real-time multi-source heterogeneous data as input data for the system operation monitoring model to obtain the output data of the system operation monitoring model; Based on the output data of the system operation monitoring model, an automated operation monitoring report corresponding to the high-altitude platform process system is generated to complete the automated operation monitoring. The system operation monitoring model, which employs a machine learning model to learn the data relationship between the historical multi-source heterogeneous data and the actual operating status, includes: An initialization strategy is used to initialize the parameters of the machine learning model, obtaining multiple parameter combinations; Based on the historical multi-source heterogeneous data and the actual operating status, obtain the fitness corresponding to each parameter combination; Based on the fitness corresponding to the parameter combination, obtain the interaction information between each parameter combination and other parameter combinations; Based on the interaction information, a fractional-order exploration strategy is used to adaptively search the parameter combination to obtain the parameter combination after adaptive exploration. Based on the interaction information, a social learning exploration strategy is used to explore the parameter combination after the adaptive exploration to obtain the parameter combination after information learning exploration. A spatial tumbling exploration strategy is used to conduct a global exploration of the parameter combination after the information learning exploration, resulting in a global exploration parameter combination; Iteratively execute the steps from obtaining fitness to global exploration until the preset training termination condition is met, and obtain the target parameter combination based on the parameter combination after the final global exploration; The model configuration parameters in the target parameter combination are used as the final parameters corresponding to the machine learning model to obtain the system operation monitoring model.

[0007] In a preferred embodiment, based on the fitness corresponding to the parameter combinations, the interaction information between each parameter combination and other parameter combinations is obtained, including: Based on the fitness corresponding to the parameter combination, the solution quality evaluation factor corresponding to the parameter combination is obtained as follows:

[0008]

[0009] in, Indicates the first t During the training process, the first i The solution quality evaluation factors corresponding to the combination of parameters This represents the fitness level corresponding to the i-th parameter combination during the t-th training process. This represents the sum of the fitness levels corresponding to all parameter combinations. Indicates the first t During the training process, the first i The fitness corresponding to each parameter combination Indicates the first t The fitness corresponding to the current worst combination of parameters in the training process. Indicates the first t The fitness of the current optimal parameter combination in the training process, where M represents the total number of parameter combinations; Based on the solution quality evaluation factors corresponding to the parameter combinations and the Euclidean distance between different parameter combinations, the interaction information between parameter combinations and other parameter combinations is obtained.

[0010] Furthermore, the interactive information is obtained in the following manner:

[0011]

[0012] in, Indicates the first t During the training process, the first i The interaction information corresponding to each parameter combination Represents the first random number between (0,1). Let ln denote the natural constant, and ln denote the logarithmic function. The interaction control coefficient is represented by t, which represents the current training iteration, and T represents the maximum training iteration. Indicates the first t During the training process, the first j A combination of parameters, Indicates the first t During the training process, the first i A combination of parameters, Indicates the first t During the training process, the first j The combination of parameters and the first i Euclidean distance between combinations of parameters This represents the base value of the interactive control coefficient.

[0013] In a preferred embodiment, based on the interaction information, a fractional-order exploration strategy is used to adaptively search the parameter combination to obtain the parameter combination after adaptive exploration, including: By utilizing the memory property of fractional calculus, combined with historical information of the current parameter combination and the interaction information, a fractional exploration formula is constructed to update the parameter combination, thereby obtaining the parameter combination after adaptive exploration.

[0014] Furthermore, the fractional-order exploration formula is as follows:

[0015] in, Indicates the first i A combination of parameters following an adaptive exploration. Indicates the order of a fraction. Indicates the first t During the training process, the first i A combination of parameters, Indicates the first t -1 training session i A combination of parameters, Indicates the first t -2 training sessions i A combination of parameters, Indicates the first t -3 training sessions i A combination of parameters, Represents the sine function. Represents pi (π). This indicates the search direction factor set to an integer. express( 0 , Random integer coefficients between ) This indicates the combination of search step sizes.

[0016] Furthermore, the fractional order The parameters are adaptively determined based on their distribution in the solution space, specifically as follows:

[0017]

[0018] in, Indicates evolutionary factors. Let represent the mean Euclidean distance between the i-th parameter combination and other parameter combinations. This represents the minimum mean value corresponding to all parameter combinations. This represents the maximum mean value corresponding to all parameter combinations.

[0019] In a preferred embodiment, based on the interaction information, a social learning exploration strategy is used to perform information learning exploration on the parameter combination after the adaptive exploration, resulting in the parameter combination after information learning exploration, including: Based on the idea of ​​particle swarm optimization, this paper introduces the individual historical optimal parameter combination and the global optimal parameter combination, and combines them with the interaction information to update the velocity and position of the parameter combination after the adaptive exploration, thus obtaining the parameter combination after information learning exploration; the social learning exploration update formula is:

[0020]

[0021] in, Indicates the first t During the training process, the first m A combination of parameters following an adaptive exploration. Indicates the first m The parameter combination after information learning and exploration. Indicates the first t +1 training session m The exploration speed corresponding to the parameter combination after adaptive exploration. Indicates the first t During the training process, the first m The exploration speed corresponding to the parameter combination after adaptive exploration. This represents the second random number between (0,1). This represents a third random number between (0,1). This represents the fourth random number between (0,1). Indicates the first learning factor. Indicates the second learning factor. Indicates the first t The current optimal parameter combination during the training process. express The historical best value, Indicates the first t During the training process, the first i The interaction information corresponding to each parameter combination.

[0022] In a preferred embodiment, a spatial tumbling exploration strategy is used to globally explore the parameter combination after the information learning exploration, resulting in a parameter combination after global exploration, including: A spatial tumbling exploration formula is constructed using a cosine function and randomly selected parameter combinations. This formula perturbs the parameter combinations obtained after information learning and exploration to escape local optima and obtain the parameter combinations after global exploration. The spatial tumbling exploration formula is as follows:

[0023]

[0024] in, Indicates the first t During the training process, the first n The parameter combination after information learning and exploration. Indicates the first n The parameter combination after a global exploration This represents the fifth random number between (0,1). Represents the cosine function. Indicates the space tumbling exploration factor. This represents the combination of parameters after random information learning and exploration. express and The Euclidean distance between them represents the roll control coefficient, t represents the current number of training iterations, and T represents the maximum number of training iterations.

[0025] In a preferred embodiment, the acquisition of multiple historical multi-source heterogeneous data corresponding to the high-altitude test platform process system and the actual operating status corresponding to the historical multi-source heterogeneous data includes: Based on a preset data sampling frequency, historical multi-source heterogeneous data is collected at N consecutive sampling time points, and the actual operating status at the N+1th sampling time point is used as the label corresponding to the historical multi-source heterogeneous data at the N consecutive sampling time points; the data is collected repeatedly to obtain multiple historical multi-source heterogeneous data and the actual operating status corresponding to each historical multi-source heterogeneous data.

[0026] In a preferred embodiment, based on the output data of the system operation monitoring model, an automated operation monitoring report corresponding to the high-altitude platform process system is generated, including: Based on the output data of the system operation monitoring model, the operating state with the highest probability value is determined as the target operating state of the high-altitude platform process system. The real-time multi-source heterogeneous data corresponding to the high-altitude platform process system and the target operating status are filled into a preset report template to obtain an automated operation monitoring report for the high-altitude platform process system.

[0027] Compared with the prior art, the present invention has at least the following beneficial effects: This invention learns the deep dynamic behavior of a system through a machine learning model. It can detect subtle anomalies in behavioral patterns before physical parameters exceed traditional safety thresholds, gaining valuable time for fault handling and effectively preventing major accidents. Simultaneously, it employs a hybrid optimization algorithm that integrates fractional-order exploration, social learning exploration, and spatial tumbling exploration for model training. This adaptively searches for optimal model configuration parameters, improving model accuracy and generalization ability, avoiding over-reliance on expert experience, and automating the monitoring process. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart of an automated operation monitoring method for a high-altitude platform process system provided in an embodiment of the present invention; Figure 2 This is a flowchart for obtaining the system operation monitoring model in an embodiment of the present invention. Detailed Implementation

[0030] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0032] like Figure 1 As shown, this embodiment of the invention provides an automated operation monitoring method for a high-altitude platform process system, comprising the following steps: S101: Collect multiple historical multi-source heterogeneous data corresponding to the high-altitude test platform process system and the actual operating status corresponding to the historical multi-source heterogeneous data; among which, the actual operating status includes normal operating status and various fault operating status.

[0033] Historical multi-source heterogeneous data can include sensor time-series data and equipment status data. Sensor time-series data can include pressure parameters (such as air source pressure, chamber pressure, loading pump inlet and outlet pressures, etc.), temperature parameters (such as intake air temperature, exhaust air temperature, cooling water temperature, lubricating oil temperature, etc.), flow parameters (such as air flow rate, fuel flow rate, cooling water flow rate, etc.), vibration parameters (such as vibration displacement, velocity, or acceleration of key equipment such as engines, pumps, and valves), speed parameters (such as engine rotor speed, pump speed, etc.), valve opening / status (such as the opening percentage or on / off status of various regulating valves, on / off valves, and manual valves), and control commands (such as setpoints and control modes from the host computer). Equipment status data can include discrete status quantities such as the start / stop status of pumps, fans, and compressors, and fault codes. Various fault operating states can be specific real fault states such as insufficient air supply pressure, excessively high cooling system water temperature, valve jamming, and surge precursors.

[0034] It is understood that the aforementioned historical multi-source heterogeneous data and the corresponding actual operating states are examples of embodiments of the present invention, and are not limited thereto. They may also include other data and states. These data may also be represented using one-hot encoding for easy identification.

[0035] In this embodiment of the invention, historical multi-source heterogeneous data and their corresponding actual operating states are collected, specifically in the following manner: Based on a preset data sampling frequency, historical multi-source heterogeneous data is collected at N consecutive sampling time points, and the actual operating state at the (N+1)th sampling time point is collected. The actual operating state at the (N+1)th sampling time point is used as the label corresponding to the historical multi-source heterogeneous data at the N consecutive sampling time points, thereby constructing input-output label pairs. This process is repeated multiple times to obtain multiple historical multi-source heterogeneous data sets and the actual operating state corresponding to each set. For example, a sampling frequency of once per second is set, and historical multi-source heterogeneous data is collected continuously at 10 sampling time points. Then, the actual operating state at the 11th sampling time point is recorded as a training sample. Repeating this process constructs a large-scale, highly consistent training dataset. This method ensures the temporal continuity and state correspondence of the data, providing high-quality learning samples for subsequent machine learning models. Compared to traditional static or asynchronous collection methods, the above collection method can more accurately capture the dynamic characteristics and state evolution patterns of the system, enhancing the model's ability to model temporal dependencies. Furthermore, the repeated data collection mechanism improves data coverage and sample diversity, effectively reduces noise interference, and enhances model robustness. This structured data collection strategy lays a solid data foundation for the subsequent construction of high-precision system operation monitoring models, and is particularly suitable for prediction scenarios.

[0036] S102: A machine learning model is used to learn the data relationship between the historical multi-source heterogeneous data and the actual operating status corresponding to the historical multi-source heterogeneous data, so as to obtain the system operation monitoring model.

[0037] In this embodiment of the invention, the machine learning model can employ a convolutional neural network (CNN) model, or a network model combining a CNN and a long short-term memory (LSTM) network. CNN is used to extract spatial features from multi-source data, while LSTM is used to process temporal dependencies in time series data. The machine learning model can learn deep correlation patterns in historical data that are difficult to describe with simple rules. It can predict potential fault trends before or in the early stages of a fault by observing subtle, correlated changes in parameters, thus achieving a fundamental shift from "post-event alarm" to "pre-event warning." By replacing human experience with machine learning models, it enables automatic and intelligent judgment of the operational status of complex high-altitude monitoring systems, greatly reducing the intensity of manual labor and the risk of subjective misjudgment.

[0038] However, model performance is highly dependent on the choice of hyperparameters, such as the number of network layers, convolutional kernel size, number of LSTM units, and learning rate. Therefore, this invention proposes a hybrid optimization algorithm to train the model, automatically searching for the optimal model configuration parameters. Figure 2 As shown, the process specifically includes the following sub-steps: S201: The machine learning model is initialized using an initialization strategy to obtain multiple parameter combinations; wherein the parameter combinations are represented in the form of vectors.

[0039] Specifically, hyperparameters are randomly initialized between their upper and lower bounds, and the initialized hyperparameters are encoded into vectors to obtain a parameter combination. This process is repeated multiple times to obtain a preset number (e.g., 50) of parameter combinations, forming an initial population. Each parameter combination represents a set of candidate model configuration parameters.

[0040] S202: Based on the historical multi-source heterogeneous data and the corresponding real operating status of the historical multi-source heterogeneous data, obtain the fitness corresponding to each parameter combination.

[0041] Specifically, the training data is divided into a training set and a validation set. For each parameter combination, the model is configured using this set of parameters, trained on the training set, and then the cross-entropy loss function or root mean square error (RMSE) is calculated on the validation set. The reciprocal of this loss value (or the reciprocal after adding a small constant) is used as the fitness of that parameter combination. For example, historical multi-source heterogeneous data can be used as input, and the actual operating state corresponding to the historical multi-source heterogeneous data can be used as the expected output. The cross-entropy loss function value or RMSE value corresponding to the parameter combination can be obtained. The cross-entropy loss function value or RMSE value is then added to a non-zero constant term (such as 0.001) and the reciprocal is taken to obtain the fitness of the parameter combination. The larger the fitness, the better the model performance of that parameter combination, and the better the position of the parameter combination in the solution space.

[0042] S203: Based on the fitness corresponding to the parameter combination, obtain the interaction information between each parameter combination and other parameter combinations.

[0043] First, the solution quality evaluation factor for each parameter combination is calculated based on fitness. Specifically, based on the fitness corresponding to the parameter combination, the solution quality evaluation factor corresponding to the parameter combination is obtained as follows:

[0044]

[0045] in, Indicates the first t During the training process, the first i The solution quality evaluation factors corresponding to the combination of parameters This represents the fitness level corresponding to the i-th parameter combination during the t-th training process. This represents the sum of the fitness levels corresponding to all parameter combinations. Indicates the first t During the training process, the first i The fitness corresponding to each parameter combination Indicates the first t The fitness corresponding to the current worst combination of parameters in the training process. Indicates the first t The fitness of the current optimal parameter combination during the training process, where M represents the total number of parameter combinations; where, With fitness Positive correlation. The higher the fitness of the parameter combination, the larger its solution quality evaluation factor.

[0046] Then, based on the solution quality evaluation factor corresponding to the parameter combination and the Euclidean distance between different parameter combinations, the interaction information between the parameter combination and other parameter combinations is obtained.

[0047] Specifically, the interactive information is obtained in the following way:

[0048]

[0049] in, Indicates the first t During the training process, the first i The interaction information corresponding to each parameter combination Represents the first random number between (0,1). Let ln denote the natural constant, and ln denote the logarithmic function. The interaction control coefficient is represented by t, which represents the current training iteration, and T represents the maximum training iteration. Indicates the first t During the training process, the first j A combination of parameters, Indicates the first t During the training process, the first i A combination of parameters, Indicates the first t During the training process, the first j The combination of parameters and the first i Euclidean distance between combinations of parameters The base value representing the interaction control coefficient can be set as a constant between 60 and 100. The interaction information reflects the degree of attraction or repulsion between different combinations of parameters.

[0050] The interaction information between all parameter vectors can be calculated based on their fitness, which can be achieved by analyzing the distribution of parameter vectors in the search space. This interaction information reflects the correlation, diversity, and guidance among different solutions in the population, providing a basis for subsequent exploration strategies.

[0051] S204: Based on the interaction information, an adaptive search is performed on the parameter combination using a fractional-order exploration strategy to obtain the parameter combination after adaptive exploration.

[0052] Specifically, the step includes: utilizing the memory property of fractional calculus, combining historical information of the current parameter combination with the interaction information, constructing a fractional exploration formula, updating the parameter combination, and obtaining the parameter combination after adaptive exploration.

[0053] Specifically, the fractional-order exploration formula is:

[0054] in, Indicates the first i A combination of parameters following an adaptive exploration. Indicates the order of a fraction. Indicates the first t During the training process, the first i A combination of parameters, Indicates the first t -1 training session i A combination of parameters, Indicates the first t -2 training sessions i A combination of parameters, Indicates the first t -3 training sessions i A combination of parameters, Represents the sine function. Represents pi (π). This indicates the search direction factor set to an integer. express( 0, Random integer coefficients between ) Indicates the combination of search steps, and and The dimensions are the same, and each element in each dimension is a random number of 0.01 or -0.01. This formula combines the memory of historical information with the random search guided by interactive information, enabling the parameter combinations to adaptively move within the solution space.

[0055] It should be noted that if it does not exist , and / or If so, the corresponding item can be set to 0.

[0056] Furthermore, the fractional order It can be adaptively determined based on the distribution of parameter combinations in the solution space, specifically:

[0057]

[0058] in, Indicates evolutionary factors. Let represent the mean Euclidean distance between the i-th parameter combination and other parameter combinations. This represents the minimum mean value corresponding to all parameter combinations. This represents the maximum mean value for all parameter combinations. When a parameter combination is far from other combinations (i.e., in a sparse region), its order is larger, enhancing exploration capabilities; when they are close to each other (in a dense region), its order is smaller, strengthening local development.

[0059] The fractional-order exploration strategy provided in this step can improve the spatial distribution of individuals, increase population diversity, overcome the drawback of the algorithm being prone to getting trapped in local optima, and combine the memory property of fractional calculus to introduce an adaptive fractional-order order. The fractional-order order is adaptively adjusted according to the location information of the individuals, further optimizing the global optimization ability of the algorithm.

[0060] S205: Based on the interaction information, a social learning exploration strategy is used to conduct information learning exploration on the parameter combination after the adaptive exploration, so as to obtain the parameter combination after information learning exploration.

[0061] Specifically, this step includes: based on the idea of ​​particle swarm optimization algorithm, introducing the individual historical optimal parameter combination and the global optimal parameter combination, and combining the interaction information to update the speed and position of the parameter combination after adaptive exploration, so as to obtain the parameter combination after information learning exploration.

[0062] An exemplary social learning exploration update formula is as follows:

[0063]

[0064] in, Indicates the first t During the training process, the first m A combination of parameters following an adaptive exploration. Indicates the first m The parameter combination after information learning and exploration. Indicates the first t +1 training session m The exploration speed corresponding to the parameter combination after adaptive exploration. Indicates the first t During the training process, the first m The exploration speed corresponding to the parameter combination after adaptive exploration. This represents the second random number between (0,1). This represents a third random number between (0,1). This represents the fourth random number between (0,1). Indicates the first learning factor. Indicates the second learning factor. Indicates the first t The current optimal parameter combination during the training process (i.e., the parameter combination with the highest fitness among all current parameter combinations). express The historical optimal value (i.e., the optimal value reached by the m-th parameter combination in the historical iteration, i.e. its individual historical optimal value). Indicates the firstt During the training process, the first i The interaction information corresponding to each parameter combination.

[0065] The social learning exploration strategy provided in this step can fully facilitate social learning, increase population diversity in the early stages of the algorithm while ensuring the training speed of the algorithm, and effectively ensure the convergence of the algorithm as it progresses, thereby improving the training capability of the algorithm.

[0066] S206: A spatial tumbling exploration strategy is used to perform a global exploration of the parameter combination after the information learning exploration, so as to obtain the parameter combination after global exploration.

[0067] Specifically, the step includes: using a cosine function and a randomly selected parameter combination to construct a space tumbling exploration formula, perturbing the parameter combination after the information learning exploration to escape local optima (i.e., helping the parameter combination escape the local optimal region of the current search space and enhance global search capabilities), and obtaining the parameter combination after global exploration.

[0068] For example, the formula for exploring space tumbling is:

[0069]

[0070] in, Indicates the first t During the training process, the first n The parameter combination after information learning and exploration. Indicates the first n The parameter combination after global exploration, where t represents the current training iteration and T represents the maximum training iteration. This represents the fifth random number between (0,1). Represents the cosine function. Indicates the space tumbling exploration factor. This represents the combination of parameters after random information learning and exploration. express and The Euclidean distance between them This represents the roll control coefficient, which can be set to an integer between 2 and 4. This formula provides a larger perturbation in the early stages of the algorithm, enhancing the global search capability; the perturbation decreases in the later stages, which is beneficial for convergence.

[0071] The spatial tumbling exploration strategy provided in this step can effectively improve the algorithm's global search capability. As the algorithm progresses, the global search capability weakens, thus avoiding stagnation or deterioration in the algorithm's training effect.

[0072] Optionally, the space tumbling exploration strategy can be improved using simulated annealing to ensure convergence in later stages. After each exploration of a parameter combination, out-of-bounds handling can be performed on hyperparameters in the parameter combination to ensure their validity.

[0073] S207: Determine whether the maximum number of training iterations has been reached. If so, obtain the target parameter combination based on the parameter combination after global exploration. Otherwise, return to the step of obtaining the fitness corresponding to the parameter combination (i.e., step S202) and continue iterating.

[0074] Steps S202 to S206 are repeated until the preset maximum number of iterations T is reached, which satisfies the preset training termination condition. In each iteration, the fitness of each parameter combination and the interaction information between parameter combinations are calculated. Higher fitness indicates better model performance for that parameter combination. Fractional-order exploration, social learning exploration, and spatial tumbling exploration are performed sequentially. Finally, the parameter combination with the highest fitness after all iterations is selected as the target parameter combination.

[0075] S208: The model configuration parameters in the target parameter combination are used as the final parameters corresponding to the machine learning model to obtain the system operation monitoring model.

[0076] Specifically, the model configuration parameters include various parameters used to define the model structure and training process, such as the number of network layers, convolutional kernel size, number of LSTM units, learning rate, and regularization coefficient. These parameters are automatically and optimally determined by the hybrid optimization algorithm described in this embodiment of the invention, enabling the model to achieve optimal performance. Using these final parameters, the machine learning model is retrained on all training data to obtain the final system operation monitoring model.

[0077] This invention innovatively proposes a model optimization algorithm that integrates fractional-order exploration, social learning exploration, and spatial tumbling exploration, significantly improving model performance and generalization ability. Traditional methods often rely on manual parameter tuning or fixed optimization strategies, which are prone to getting trapped in local optima and are inefficient. This invention, however, innovatively integrates fractional-order exploration, social learning exploration, and spatial tumbling exploration strategies based on the interaction information between fitness evaluation parameters. This achieves adaptive parameter space, information guidance, and global search, effectively avoiding premature convergence and enhancing global optimization capabilities. Furthermore, this mechanism iteratively optimizes until the maximum number of training iterations, ensuring that the parameter combinations gradually approach the optimal solution, thereby obtaining a high-precision and robust system operation monitoring model. Compared to existing technologies, this invention has significant advantages in model training efficiency, prediction accuracy, and adaptability to complex scenarios, and is particularly suitable for complex system monitoring tasks under multi-source heterogeneous data.

[0078] S103: Collect real-time multi-source heterogeneous data corresponding to the high-altitude platform process system, and use the real-time multi-source heterogeneous data as input data for the system operation monitoring model, and obtain the output data of the system operation monitoring model.

[0079] Specifically, the data structure of real-time multi-source heterogeneous data is the same as that of historical multi-source heterogeneous data, also including various sensor time-series data and equipment status data, ensuring that the system operation monitoring model can accurately identify the data. This real-time data is input into the trained system operation monitoring model, which outputs a probability vector representing the probability that the current system is in normal operation or various fault states, thus obtaining the output data of the system operation monitoring model.

[0080] S104: Based on the output data of the system operation monitoring model, generate an automated operation monitoring report for the high-altitude platform process system to complete the automated operation monitoring.

[0081] Specifically, this step includes: based on the output data of the system operation monitoring model, determining the operating state with the highest probability value as the target operating state corresponding to the high-altitude test station process system; filling the real-time multi-source heterogeneous data corresponding to the high-altitude test station process system and the target operating state into a preset report template to obtain an automated operation monitoring report corresponding to the high-altitude test station process system.

[0082] For example, if the probability of "normal operation" is the highest among the probabilities output by the model, then the target operating state is determined to be normal operation; if the probability of "insufficient gas pressure fault" is the highest, then the target state is determined to be that fault. Then, real-time multi-source heterogeneous data (which may include the values ​​of key parameters) and the determined target state are filled into a preset report template to generate a monitoring report containing information such as timestamp, operating status, key data, and warning level.

[0083] Optionally, when the real-time operating status of the high-altitude platform process system is abnormal, the automated operation monitoring report is transmitted to the device designated by the staff (such as a mobile phone or computer) and to the blockchain for archiving, so as to ensure the authenticity and traceability of the fault data by utilizing the immutability of the blockchain.

[0084] This completes the entire automated operation and monitoring process.

[0085] The present invention provides an automated operation monitoring method for a high-altitude test bench process system. Through accurate early warning and rapid response, it can effectively avoid or mitigate losses caused by malfunctions, ensure the safety of the high-altitude test bench equipment and the engine under test, and improve the success rate of the test and the reliability of equipment operation.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automated run monitoring of a high altitude table process system, characterized in that, Includes the following steps: Collect multiple historical multi-source heterogeneous data corresponding to the high-altitude test platform process system and the actual operating status corresponding to the historical multi-source heterogeneous data; wherein, the actual operating status includes normal operating status and various fault operating status; A machine learning model is used to learn the data relationship between the historical multi-source heterogeneous data and the actual operating status to obtain a system operation monitoring model; Collect real-time multi-source heterogeneous data corresponding to the high-altitude platform process system, and use the real-time multi-source heterogeneous data as input data for the system operation monitoring model to obtain the output data of the system operation monitoring model; Based on the output data of the system operation monitoring model, an automated operation monitoring report corresponding to the high-altitude platform process system is generated to complete the automated operation monitoring. The system operation monitoring model, which employs a machine learning model to learn the data relationship between the historical multi-source heterogeneous data and the actual operating status, includes: The machine learning model is initialized using an initialization strategy to obtain multiple parameter combinations; Based on the historical multi-source heterogeneous data and the actual operating status, obtain the fitness corresponding to each parameter combination; Based on the fitness corresponding to the parameter combination, obtain the interaction information between each parameter combination and other parameter combinations; Based on the interaction information, a fractional-order exploration strategy is used to adaptively search the parameter combination to obtain the parameter combination after adaptive exploration. Based on the interaction information, a social learning exploration strategy is used to explore the parameter combination after the adaptive exploration to obtain the parameter combination after information learning exploration. A spatial tumbling exploration strategy is used to conduct a global exploration of the parameter combination after the information learning exploration, resulting in a global exploration parameter combination; Iteratively execute the steps from obtaining fitness to global exploration until the preset training termination condition is met, and obtain the target parameter combination based on the parameter combination after the final global exploration; The model configuration parameters in the target parameter combination are used as the final parameters corresponding to the machine learning model to obtain the system operation monitoring model.

2. The method for monitoring automation of operation of a high altitude table process system according to claim 1, wherein, Based on the fitness corresponding to the parameter combinations, obtain the interaction information between each parameter combination and other parameter combinations, including: Based on the fitness corresponding to the parameter combination, the solution quality evaluation factor corresponding to the parameter combination is obtained as follows: in, Indicates the first t During the training process, the first i The solution quality evaluation factors corresponding to the combination of parameters This represents the fitness level corresponding to the i-th parameter combination during the t-th training process. This represents the sum of the fitness levels corresponding to all parameter combinations. Indicates the first t During the training process, the first i The fitness corresponding to each parameter combination Indicates the first t The fitness corresponding to the current worst combination of parameters in the training process. Indicates the first t The fitness of the current optimal parameter combination in the training process, where M represents the total number of parameter combinations; Based on the solution quality evaluation factors corresponding to the parameter combinations and the Euclidean distance between different parameter combinations, the interaction information between parameter combinations and other parameter combinations is obtained.

3. The automated operation monitoring method for a high-altitude platform process system according to claim 2, characterized in that, The interactive information is obtained through the following methods: in, Indicates the first t During the training process, the first i The interaction information corresponding to each parameter combination Represents the first random number between (0,1). Let ln denote the natural constant, and ln denote the logarithmic function. The interaction control coefficient is represented by t, which represents the current training iteration, and T represents the maximum training iteration. Indicates the first t During the training process, the first j A combination of parameters, Indicates the first t During the training process, the first i A combination of parameters, Indicates the first t During the training process, the first j The combination of parameters and the first i Euclidean distance between combinations of parameters This represents the base value of the interactive control coefficient.

4. The automated operation monitoring method for a high-altitude platform process system according to claim 1, characterized in that, Based on the interaction information, a fractional-order exploration strategy is used to adaptively search the parameter combination to obtain the parameter combination after adaptive exploration, including: By utilizing the memory property of fractional calculus, combined with historical information of the current parameter combination and the interaction information, a fractional exploration formula is constructed to update the parameter combination, thereby obtaining the parameter combination after adaptive exploration.

5. The automated operation monitoring method for a high-altitude platform process system according to claim 4, characterized in that, The fractional-order exploration formula is as follows: in, Indicates the first i A combination of parameters following an adaptive exploration. Indicates the order of a fraction. Indicates the first t During the training process, the first i A combination of parameters, Indicates the first t -1 training session i A combination of parameters, Indicates the first t -2 training sessions i A combination of parameters, Indicates the first t -3 training sessions i A combination of parameters, Represents the sine function. Represents pi (π). This indicates the search direction factor set to an integer. express( 0 , Random integer coefficients between ) This indicates the combination of search step sizes.

6. The automated operation monitoring method for a high-altitude platform process system according to claim 5, characterized in that, The fractional order The parameters are adaptively determined based on their distribution in the solution space, specifically as follows: in, Indicates evolutionary factors. Let represent the mean Euclidean distance between the i-th parameter combination and other parameter combinations. This represents the minimum mean value corresponding to all parameter combinations. This represents the maximum mean value corresponding to all parameter combinations.

7. The automated operation monitoring method for a high-altitude platform process system according to claim 1, characterized in that, Based on the interaction information, a social learning exploration strategy is used to explore the parameter combination after the adaptive exploration, resulting in the parameter combination after information learning exploration, including: Based on the idea of ​​particle swarm optimization, this paper introduces the individual historical optimal parameter combination and the global optimal parameter combination, and combines them with the interaction information to update the velocity and position of the parameter combination after the adaptive exploration, thus obtaining the parameter combination after information learning exploration; the social learning exploration update formula is: in, Indicates the first t During the training process, the first m A combination of parameters following an adaptive exploration. Indicates the first m The parameter combination after information learning and exploration. Indicates the first t +1 training session m The exploration speed corresponding to the parameter combination after adaptive exploration. Indicates the first t During the training process, the first m The exploration speed corresponding to the parameter combination after adaptive exploration. This represents the second random number between (0,1). This represents a third random number between (0,1). This represents the fourth random number between (0,1). Indicates the first learning factor. Indicates the second learning factor. Indicates the first t The current optimal parameter combination during the training process. express The historical best value, Indicates the first t During the training process, the first i The interaction information corresponding to each parameter combination.

8. The automated operation monitoring method for a high-altitude platform process system according to claim 1, characterized in that, A spatial tumbling exploration strategy is employed to globally explore the parameter combinations after the information learning exploration, resulting in the following parameter combinations after global exploration: A spatial tumbling exploration formula is constructed using a cosine function and randomly selected parameter combinations. This formula perturbs the parameter combinations obtained after information learning and exploration to escape local optima and obtain the parameter combinations after global exploration. The spatial tumbling exploration formula is as follows: in, Indicates the first t During the training process, the first n The parameter combination after information learning and exploration. Indicates the first n The parameter combination after a global exploration This represents the fifth random number between (0,1). Represents the cosine function. Indicates the space tumbling exploration factor. This represents the combination of parameters after random information learning and exploration. express and The Euclidean distance between them represents the roll control coefficient, t represents the current number of training iterations, and T represents the maximum number of training iterations.

9. The automated operation monitoring method for a high-altitude platform process system according to claim 1, characterized in that, Collect multiple historical multi-source heterogeneous data corresponding to the high-altitude platform process system, as well as the actual operating status corresponding to the historical multi-source heterogeneous data, including: Based on a preset data sampling frequency, historical multi-source heterogeneous data is collected at N consecutive sampling time points, and the actual operating status at the N+1th sampling time point is used as the label corresponding to the historical multi-source heterogeneous data at the N consecutive sampling time points; the data is collected repeatedly to obtain multiple historical multi-source heterogeneous data and the actual operating status corresponding to each historical multi-source heterogeneous data.

10. The automated operation monitoring method for a high-altitude platform process system according to claim 1, characterized in that, Based on the output data of the system operation monitoring model, an automated operation monitoring report corresponding to the high-altitude platform process system is generated, including: Based on the output data of the system operation monitoring model, the operating state with the highest probability value is determined as the target operating state of the high-altitude platform process system. The real-time multi-source heterogeneous data corresponding to the high-altitude platform process system and the target operating status are filled into a preset report template to obtain an automated operation monitoring report for the high-altitude platform process system.

Images