Deep learning-based attitude stabilization control method for offshore rocket launch platform

Abstract

This invention discloses a deep learning-based attitude stabilization control method for a marine rocket launch platform, belonging to the field of deep learning technology. It collects attitude angle, roll rate, and environmental disturbance force data from the platform's historical operation, and determines the operating state based on preset intervals. It statistically analyzes historical operating node times and establishes a mapping relationship between time and operating state. By analyzing state transition events of adjacent nodes, it generates an operating state transition matrix and calculates the transition probability and error transition probability. Based on the highest transition probability, it predicts the next operating state, dynamically corrects the target attitude angle threshold based on the error transition probability, and selects to execute a control strategy or switch directly based on a preset error threshold. This achieves predictive-driven adaptive attitude adjustment, improving the robustness and response efficiency of platform attitude control, reducing instability risk and control energy consumption, extending equipment lifespan, and significantly improving the safety and reliability of the rocket launch platform.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, specifically to a deep learning-based attitude stabilization control method for a marine rocket launch platform. Background Technology

[0002] With the rapid rise of global commercial spaceflight, traditional land-based launch sites are increasingly unable to meet the demands for frequent, low-cost launches. Sea-based launches, due to their mobility, low dependence on ground facilities, and reduced risks from densely populated areas, have become a focus of research in recent years. However, the extreme uncertainty of the marine environment makes platform attitude stability control a core challenge restricting the reliability of sea-based rocket launches. Most existing attitude control methods are based on traditional control theory. While these methods perform well in stable environments and with limited disturbances, they are prone to problems such as control delays, insufficient robustness, and decreased prediction accuracy when facing complex nonlinear disturbances and rapidly changing sea states. Furthermore, existing control strategies generally rely on pre-set physical models or empirical formulas. When environmental parameters exceed expected ranges, the model's adaptability is insufficient, often leading to control failure or error accumulation, thereby reducing the safety and stability of the launch process.

[0003] To address the aforementioned shortcomings, academia and industry have gradually introduced intelligent algorithms such as deep learning in recent years to improve the adaptability and predictive capabilities of control systems. However, existing research has largely focused on general ship dynamic positioning or stability control of marine engineering platforms, lacking systematic methods for the specific application scenario of offshore rocket launch platforms. Existing technologies have significant limitations in the accuracy of operational state identification, the refinement of state transition law modeling, and real-time decision-making capabilities under multiple disturbance conditions. For example, traditional methods struggle to fully capture the complex temporal correlation between attitude angles, roll angular velocity, and environmental disturbance forces, failing to form a dynamically evolving state transition matrix, resulting in significant prediction deviations. Furthermore, in designing trigger conditions for attitude control strategies, existing methods often rely on single threshold judgments, lacking dynamic adjustment mechanisms based on error transition probabilities, thus resulting in insufficient response under sudden, strong disturbance conditions and affecting control performance. Summary of the Invention

[0004] The purpose of this invention is to provide a deep learning-based attitude stabilization control method for marine rocket launch platforms to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A deep learning-based attitude stabilization control method for a marine rocket launch platform includes the following steps: Step S1: Acquire historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the marine rocket launch platform during its historical operation; determine the operating state of the launch platform, including a stable state, a moderate disturbance state, and a strong disturbance state; Step S2: Statistically analyze all historical operating node times of the launch platform and generate a set of historical operating node times; map the operating states to the historical operating node times; Step S3: Mark operating state transition events according to the mapping relationship; identify all operating state transition events and generate an operating state transition matrix; calculate the operating state transition probability of any operating state transition event; calculate the error transition probability between operating state transition probabilities; Step S4: Extract all operating state transition probabilities, and take the operating state with the highest probability as the next operating state for preliminary prediction; calculate the target attitude angle threshold; preset an error threshold; if the total error transition probability is greater than or equal to the error threshold, execute the control strategy; if the total error transition probability is less than the error threshold, directly switch to the target attitude angle threshold.

[0007] As a preferred embodiment of the deep learning-based attitude stabilization control method for a marine rocket launch platform described in this invention, historical operating state parameters of the marine rocket launch platform during its historical operation are obtained through sensor and data processing analysis technology. These historical operating state parameters include historical attitude angle data, historical roll rate data, and historical environmental disturbance force data. Based on these historical attitude angle data, historical roll rate data, and historical environmental disturbance force data, the operating state of the launch platform is determined, as follows:

[0008] Preset attitude angle data range, roll rate data range, and environmental disturbance force data range;

[0009] If the historical attitude angle data is less than or equal to the minimum value of the attitude angle data range, the historical roll rate data is less than or equal to the minimum value of the roll rate data range, and the historical environmental disturbance force data is less than or equal to the minimum value of the environmental disturbance force data range, then the operating state is determined to be a stable state.

[0010] If the historical attitude angle data is within the attitude angle data range, the historical roll rate data is within the roll rate data range, and the historical environmental disturbance force data is within the environmental disturbance force data range, then the operating state is determined to be a moderate disturbance state.

[0011] If the historical attitude angle data is greater than or equal to the maximum value of the attitude angle data range, the historical roll rate data is greater than or equal to the maximum value of the roll rate data range, and the historical environmental disturbance force data is greater than or equal to the maximum value of the environmental disturbance force data range, then the operating state is determined to be a strong disturbance state.

[0012] As a preferred embodiment of the deep learning-based attitude stabilization control method for a marine rocket launch platform described in this invention, the historical operation time of the launch platform is statistically analyzed, and a set of historical operation time is generated, denoted as... ,in, Let T represent the time of the t-th historical running node, and T represent the total number of historical running node times; the stable state, the moderately disturbed state, and the strongly disturbed state are respectively labeled as... , and Based on the historical runtime time set, a time-to-running-state mapping is established between the running state and the historical runtime time. Each historical runtime time corresponds to one running state. ,and .

[0013] As a preferred embodiment of the deep learning-based attitude stabilization control method for a marine rocket launch platform described in this invention, the operational state transition events are analyzed according to the mapping relationship. If an operational state transition occurs between two adjacent historical operational time nodes, it is marked as an operational state transition event. The operational state transition is represented as... ,in, And i, j ≠ v; identify all runtime state transition events and generate a runtime state transition matrix, denoted as:

[0014]

[0015] in, Indicates the running status Transition to running state ,Right now ;

[0016] Based on the operational state transition matrix and the historical operational node time set, the number of each type of operational state transition event is counted, denoted as . Based on the number of each type of runtime state transition event, calculate the runtime state transition probability of any given event. The specific calculation formula is as follows:

[0017]

[0018] in, Indicates by The probability of state transition for a state transition event;

[0019] Calculate running status Transition to running state The probability of state transition and the running state Transition to running state The error transition probability between the operating state transition probabilities is calculated using the following formula:

[0020]

[0021] in, Indicates the error transition probability. Indicates the preset operating state The perturbation weights, This indicates the preset error term. Indicates the running status to running status The probability of state transition for a state transition event.

[0022] As a preferred embodiment of the deep learning-based attitude stabilization control method for a marine rocket launch platform described in this invention, if the current operating state is... Extract all running state transition probabilities The operating state with the highest probability The next operational state, as initially predicted, is as follows: ;

[0023] According to the next running state and error transition probability Adjust the target attitude angle threshold as follows:

[0024]

[0025] in, Indicates the target attitude angle threshold. Indicates running status The corresponding physical limit value of the attitude angle, This represents the preset correction coefficient for the current attitude deviation. Indicates running status The corresponding attitude angle, Indicates the current actual attitude angle. This indicates the preset adjustment coefficient. Indicates running status The total error transition probability to all other non-self states;

[0026] A preset error threshold is set if the total error transition probability is... If the error threshold is greater than or equal to the target attitude angle threshold, then the target attitude angle threshold is used. The control policy is executed as follows:

[0027]

[0028] in, Indicates the attitude angle adjustment step size. This indicates the preset maximum attitude adjustment step size. This represents the preset attitude deviation threshold. This indicates the preset basic posture adjustment step size;

[0029] If the total error transfer probability If the value is less than the error threshold, then directly switch to the target attitude angle threshold. .

[0030] This is a deep learning-based attitude stabilization control system for a marine rocket launch platform. The system includes: a data acquisition and state judgment module, a mapping module, a matrix construction and error calculation module, and a threshold calculation and control module.

[0031] The data acquisition and status judgment module acquires historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the sea-based rocket launch platform during its historical operation; and judges the operating status of the launch platform, which includes a stable state, a moderate disturbance state, and a strong disturbance state.

[0032] The mapping module: counts all historical operating node times of the launch platform and generates a set of historical operating node times; maps the operating status to the historical operating node times.

[0033] The matrix construction and error calculation module: marks the running state transition events according to the mapping relationship; identifies all running state transition events and generates a running state transition matrix; calculates the running state transition probability of any running state transition event; and calculates the error transition probability between the running state transition probabilities.

[0034] The threshold calculation and control module extracts all operating state transition probabilities and uses the operating state with the highest probability as the next operating state for preliminary prediction; calculates the target attitude angle threshold; presets an error threshold; if the total error transition probability is greater than or equal to the error threshold, the control strategy is executed; if the total error transition probability is less than the error threshold, the system directly switches to the target attitude angle threshold.

[0035] Furthermore, the data acquisition and status determination module includes a data acquisition unit and a status determination unit;

[0036] The data acquisition unit acquires historical operational status parameters of the sea-based rocket launch platform during its historical operation through sensors and data processing and analysis technology. These historical operational status parameters include historical attitude angle data, historical roll rate data, and historical environmental disturbance force data. Based on the historical attitude angle data, historical roll rate data, and historical environmental disturbance force data...

[0037] The state determination unit determines the operating state of the launch platform as follows: It presets an attitude angle data range, a roll rate data range, and an environmental interference force data range. If the historical attitude angle data is less than or equal to the minimum value of the attitude angle data range, the historical roll rate data is less than or equal to the minimum value of the roll rate data range, and the historical environmental interference force data is less than or equal to the minimum value of the environmental interference force data range, then the operating state is determined to be a stable state. If the historical attitude angle data is within the attitude angle data range, the historical roll rate data is within the roll rate data range, and the historical environmental interference force data is within the environmental interference force data range, then the operating state is determined to be a moderate disturbance state. If the historical attitude angle data is greater than or equal to the maximum value of the attitude angle data range, the historical roll rate data is greater than or equal to the maximum value of the roll rate data range, and the historical environmental interference force data is greater than or equal to the maximum value of the environmental interference force data range, then the operating state is determined to be a strong disturbance state.

[0038] Furthermore, the mapping module includes a mapping unit;

[0039] The mapping unit: counts all historical operating node times of the launch platform and generates a set of historical operating node times; based on the set of historical operating node times, maps the operating state to the historical operating node times in a time-to-operating state correspondence, wherein one historical operating node time corresponds to one operating state.

[0040] Furthermore, the matrix construction and error calculation module includes a matrix construction unit and an error calculation unit;

[0041] The matrix construction unit: sorts out the running state transition events according to the mapping relationship; if a running state transition occurs between two adjacent historical running node times, it is marked as a running state transition event; identifies all running state transition events and generates a running state transition matrix;

[0042] The error calculation unit: based on the running state transition matrix and the historical running node time set, counts the number of each running state transition event; based on the number of each running state transition event, calculates the running state transition probability of any running state transition event; and calculates the error transition probability between the running state transition probabilities.

[0043] Furthermore, the threshold calculation and control module includes a threshold calculation unit and a control unit;

[0044] The threshold calculation unit: extracts all operating state transition probabilities, takes the operating state with the highest probability as the next operating state for preliminary prediction; and adjusts the target attitude angle threshold according to the next operating state and the error transition probability.

[0045] The control unit has a preset error threshold. If the total error transition probability is greater than or equal to the error threshold, the control strategy is executed based on the target attitude angle threshold. If the total error transition probability is less than the error threshold, the control unit directly switches to the target attitude angle threshold.

[0046] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The deep learning-based attitude stabilization control method for a marine rocket launch platform provided by this invention acquires historical attitude angle, roll rate, and environmental disturbance force data, and combines this with preset intervals to determine the operating state, thus achieving precise classification of platform operating conditions and laying a data foundation for subsequent prediction and control. By statistically analyzing historical operating node times and establishing a mapping relationship between time and operating state, time-series modeling of operating state is achieved, enabling the identification of the regularity and time dependence of disturbance occurrences, thereby improving the sensitivity and accuracy of state prediction. By marking and statistically analyzing operating state transition events, generating an operating state transition matrix, and calculating transition probabilities and error transition probabilities, a quantitative description of the platform state evolution trend and uncertainty is achieved, enabling the control system to possess early warning and risk trade-off capabilities. By extracting the highest transition probability to predict the next operating state, and dynamically correcting the target attitude angle threshold based on the error transition probability, and then selecting to execute a control strategy or directly switch according to the error threshold, prediction-driven adaptive attitude adjustment is achieved, improving response speed and energy efficiency while ensuring safety margins. Overall, this method forms a closed-loop framework of data acquisition, time-series mapping, probabilistic modeling, and adaptive execution. It not only enhances the robustness and adaptability of the sea-based rocket launch platform in complex disturbance environments and reduces the risk of attitude instability and launch failure, but also reduces wear and tear on actuators and control energy consumption, thereby significantly improving the safety, stability, and mission reliability of the platform operation. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0048] Figure 1 This is a schematic diagram of the steps of the deep learning-based attitude stabilization control method for a marine rocket launch platform according to the present invention.

[0049] Figure 2This is a schematic diagram of the attitude stabilization control system for a marine rocket launch platform based on deep learning, as described in this invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Please see Figure 1 In this first embodiment: a deep learning-based attitude stabilization control method for a marine rocket launch platform is provided, which includes the following steps:

[0052] Step S1: Obtain historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the sea-based rocket launch platform during its historical operation; determine the operating status of the launch platform, which includes a stable state, a moderate disturbance state, and a strong disturbance state.

[0053] Specifically, by using sensor and data processing analysis technologies, historical operational status parameters of the sea-based rocket launch platform are acquired during its historical operation. These historical operational status parameters include historical attitude angle data, historical roll rate data, and historical environmental disturbance force data. Based on these historical attitude angle data, historical roll rate data, and historical environmental disturbance force data, the operational status of the launch platform is determined, as follows:

[0054] Preset attitude angle data range, roll rate data range, and environmental disturbance force data range;

[0055] If the historical attitude angle data is less than or equal to the minimum value of the attitude angle data range, the historical roll rate data is less than or equal to the minimum value of the roll rate data range, and the historical environmental disturbance force data is less than or equal to the minimum value of the environmental disturbance force data range, then the operating state is determined to be a stable state.

[0056] If the historical attitude angle data is within the attitude angle data range, the historical roll rate data is within the roll rate data range, and the historical environmental disturbance force data is within the environmental disturbance force data range, then the operating state is determined to be a moderate disturbance state.

[0057] If the historical attitude angle data is greater than or equal to the maximum value of the attitude angle data range, the historical roll rate data is greater than or equal to the maximum value of the roll rate data range, and the historical environmental disturbance force data is greater than or equal to the maximum value of the environmental disturbance force data range, then the operating state is determined to be a strong disturbance state.

[0058] It should be noted that by acquiring historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the sea-based rocket launch platform, and judging its operational status (stable / moderate disturbance / strong disturbance) accordingly, quantitative classification and characterization of the platform's operational behavior were achieved. This provides reliable historical samples and labels for subsequent construction of time-series models, training of deep learning networks, and setting of state-related control strategies, enabling the model to distinguish dynamic characteristics under different disturbance levels. It significantly reduces the error rate of state determination and the probability of false control triggering, providing a robust data foundation for subsequent prediction and graded control, thereby improving the platform's predictive ability and launch reliability under different sea conditions.

[0059] Step S2: Calculate the historical operating node times of the launch platform and generate a set of historical operating node times; map the operating status to the historical operating node times.

[0060] Specifically, the historical operation time of the launch platform is statistically analyzed, and a set of historical operation time is generated, denoted as . ,in, Let T represent the time of the t-th historical running node, and T represent the total number of historical running node times; the stable state, the moderately disturbed state, and the strongly disturbed state are respectively labeled as... , and Based on the historical runtime time set, a time-to-running-state mapping is established between the running state and the historical runtime time. Each historical runtime time corresponds to one running state. ,and .

[0061] It should be noted that by statistically analyzing the times of all historical operating nodes and establishing a time-to-operating state mapping, the discrete historical observations are stored and indexed in a structured manner according to time series. This facilitates the analysis of the time distribution, dwell time, and temporal correlation of state occurrences (e.g., day / night / seasonal / meteorological correlation patterns), providing the necessary temporal context for calculating state transition frequency, identifying periodic disturbance patterns, and generating training / validation sets (time division). It also improves the predictive sensitivity to "when and what kind of disturbance occurs," reduces prediction bias caused by ignoring time correlation, and supports adaptive thresholds and control strategies based on time windows, thereby improving the timeliness and accuracy of control strategies under time-varying sea states.

[0062] Step S3: Mark the running state transition events according to the mapping relationship; identify all running state transition events and generate a running state transition matrix; calculate the running state transition probability of any running state transition event; calculate the error transition probability between running state transition probabilities.

[0063] Specifically, based on the mapping relationship, the runtime state transition events are sorted out. If a runtime state transition occurs between two adjacent historical runtime nodes, it is marked as a runtime state transition event. The runtime state transition is represented as... ,in, And i, j ≠ v; identify all runtime state transition events and generate a runtime state transition matrix, denoted as:

[0064]

[0065] in, Indicates the running status Transition to running state ,Right now ;

[0066] Based on the operational state transition matrix and the historical operational node time set, the number of each type of operational state transition event is counted, denoted as . Based on the number of each type of runtime state transition event, calculate the runtime state transition probability of any given event. The specific calculation formula is as follows:

[0067]

[0068] in, Indicates by The probability of state transition for a state transition event;

[0069] Calculate running status Transition to running state The probability of state transition and the running state Transition to running state The error transition probability between the operating state transition probabilities is calculated using the following formula:

[0070]

[0071] in, Indicates the error transition probability. Indicates the preset operating state The perturbation weights, This indicates the preset error term. Indicates the running status to running status The probability of state transition for a state transition event.

[0072] It should be noted that by marking and identifying the operational state transitions of adjacent historical nodes and generating operational state transition matrices and various transition probabilities, and then calculating error transition probabilities, probabilistic modeling and uncertainty quantification of the platform's operational state evolution are achieved. This enables the system to describe the trend and confidence level of "evolution from one state to another" using quantitative probabilities, and to make model bias, disturbance weights, and uncertainties explicit through error transition probabilities, which can be used to weigh risks in prediction and control decisions. This step gives the system early warning and risk assessment capabilities—when the probability of a certain type of high-risk transition increases, a more conservative / more aggressive control strategy can be triggered in advance to reduce the risk of attitude instability caused by sudden disturbances and improve the interpretability and provability of control decisions (facilitating post-analysis and responsibility attribution).

[0073] Step S4: Extract all running state transition probabilities, and take the running state with the highest probability as the next running state for preliminary prediction; calculate the target attitude angle threshold; preset the error threshold. If the total error transition probability is greater than or equal to the error threshold, then execute the control strategy; if the total error transition probability is less than the error threshold, then directly switch to the target attitude angle threshold.

[0074] Specifically, if the current running status is Extract all running state transition probabilities The operating state with the highest probability The next operational state, as initially predicted, is as follows: According to the next running state and error transition probability Adjust the target attitude angle threshold as follows:

[0075]

[0076] in, Indicates the target attitude angle threshold. Indicates running status The corresponding physical limit value of the attitude angle, This represents the preset correction coefficient for the current attitude deviation. Indicates running status The corresponding attitude angle, Indicates the current actual attitude angle. This indicates the preset adjustment coefficient. Indicates running status The total error transition probability to all other non-self states;

[0077] A preset error threshold is set if the total error transition probability is... If the error threshold is greater than or equal to the target attitude angle threshold, then the target attitude angle threshold is used. The control policy is executed as follows:

[0078]

[0079] in, Indicates the attitude angle adjustment step size. This indicates the preset maximum attitude adjustment step size. This represents the preset attitude deviation threshold. This indicates the preset basic posture adjustment step size;

[0080] If the total error transfer probability If the value is less than the error threshold, then directly switch to the target attitude angle threshold. .

[0081] It should be noted that by extracting the next running state with the maximum transition probability, dynamically adjusting the target attitude angle threshold based on this state and the error transition probability, and selecting "execute control strategy" or "directly switch to target threshold" according to the total error transition probability and the preset threshold, prediction-driven and asymmetric robust adaptive control decision is achieved.

[0082] When the prediction shows a high-risk transfer and a large probability of error, the system adopts a step-by-step adjustment (limited by the maximum step size and the basic step size) to avoid overshoot or actuator saturation caused by a large one-time action; when the risk is low, direct switching is allowed to shorten the stabilization time.

[0083] This step optimizes response speed and energy consumption while ensuring safety margins, reduces the number of control activations and actuator wear (because fine adjustments are only made when necessary), and reduces attitude overshoot and steady-state error, thereby increasing the probability of attitude compliance within the launch window and the overall launch success rate.

[0084] Please see Figure 2 In this second embodiment: a deep learning-based attitude stabilization control system for a marine rocket launch platform is provided. The system includes: a data acquisition and state judgment module, a mapping module, a matrix construction and error calculation module, and a threshold calculation and control module.

[0085] The data acquisition and status judgment module acquires historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the sea-based rocket launch platform during its historical operation; and judges the operating status of the launch platform, which includes a stable state, a moderate disturbance state, and a strong disturbance state.

[0086] The mapping module: counts all historical operating node times of the launch platform and generates a set of historical operating node times; maps the operating status to the historical operating node times.

[0087] The matrix construction and error calculation module: marks the running state transition events according to the mapping relationship; identifies all running state transition events and generates a running state transition matrix; calculates the running state transition probability of any running state transition event; and calculates the error transition probability between the running state transition probabilities.

[0088] The threshold calculation and control module extracts all operating state transition probabilities and uses the operating state with the highest probability as the next operating state for preliminary prediction; calculates the target attitude angle threshold; presets an error threshold; if the total error transition probability is greater than or equal to the error threshold, the control strategy is executed; if the total error transition probability is less than the error threshold, the system directly switches to the target attitude angle threshold.

[0089] Furthermore, the data acquisition and status determination module includes a data acquisition unit and a status determination unit;

[0090] The data acquisition unit acquires historical operational status parameters of the sea-based rocket launch platform during its historical operation through sensors and data processing and analysis technology. These historical operational status parameters include historical attitude angle data, historical roll rate data, and historical environmental disturbance force data. Based on the historical attitude angle data, historical roll rate data, and historical environmental disturbance force data...

[0091] The state determination unit determines the operating state of the launch platform as follows: It presets an attitude angle data range, a roll rate data range, and an environmental interference force data range. If the historical attitude angle data is less than or equal to the minimum value of the attitude angle data range, the historical roll rate data is less than or equal to the minimum value of the roll rate data range, and the historical environmental interference force data is less than or equal to the minimum value of the environmental interference force data range, then the operating state is determined to be a stable state. If the historical attitude angle data is within the attitude angle data range, the historical roll rate data is within the roll rate data range, and the historical environmental interference force data is within the environmental interference force data range, then the operating state is determined to be a moderate disturbance state. If the historical attitude angle data is greater than or equal to the maximum value of the attitude angle data range, the historical roll rate data is greater than or equal to the maximum value of the roll rate data range, and the historical environmental interference force data is greater than or equal to the maximum value of the environmental interference force data range, then the operating state is determined to be a strong disturbance state.

[0092] Furthermore, the mapping module includes a mapping unit;

[0093] The mapping unit: counts all historical operating node times of the launch platform and generates a set of historical operating node times; based on the set of historical operating node times, maps the operating state to the historical operating node times in a time-to-operating state correspondence, wherein one historical operating node time corresponds to one operating state.

[0094] Furthermore, the matrix construction and error calculation module includes a matrix construction unit and an error calculation unit;

[0095] The matrix construction unit: sorts out the running state transition events according to the mapping relationship; if a running state transition occurs between two adjacent historical running node times, it is marked as a running state transition event; identifies all running state transition events and generates a running state transition matrix;

[0096] The error calculation unit: based on the running state transition matrix and the historical running node time set, counts the number of each running state transition event; based on the number of each running state transition event, calculates the running state transition probability of any running state transition event; and calculates the error transition probability between the running state transition probabilities.

[0097] Furthermore, the threshold calculation and control module includes a threshold calculation unit and a control unit;

[0098] The threshold calculation unit: extracts all operating state transition probabilities, takes the operating state with the highest probability as the next operating state for preliminary prediction; and adjusts the target attitude angle threshold according to the next operating state and the error transition probability.

[0099] The control unit has a preset error threshold. If the total error transition probability is greater than or equal to the error threshold, the control strategy is executed based on the target attitude angle threshold. If the total error transition probability is less than the error threshold, the control unit directly switches to the target attitude angle threshold.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. 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 attitude stabilization control of a sea-based rocket launch platform based on deep learning, characterized in that, The method includes the following steps: Step S1: Obtain historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the sea-based rocket launch platform during its historical operation; determine the operating status of the launch platform, which includes a stable state, a moderate disturbance state, and a strong disturbance state; Step S2: Count all historical operating node times of the launch platform and generate a set of historical operating node times; map the operating status to the historical operating node times; Step S3: Mark the running state transition events according to the mapping relationship; identify all running state transition events and generate a running state transition matrix; calculate the running state transition probability of any running state transition event; calculate the error transition probability between running state transition probabilities; Step S4: Extract all running state transition probabilities, and take the running state with the highest probability as the next running state for preliminary prediction; calculate the target attitude angle threshold; preset the error threshold. If the total error transition probability is greater than or equal to the error threshold, then execute the control strategy; if the total error transition probability is less than the error threshold, then directly switch to the target attitude angle threshold. The specific implementation process of step S2 includes: The historical operation time of the launch platform is statistically analyzed, and a set of historical operation time is generated, denoted as . ,in, Let T represent the time of the t-th historical running node, and T represent the total number of historical running node times; the stable state, the moderately disturbed state, and the strongly disturbed state are respectively labeled as... , and Based on the historical runtime time set, a time-to-running-state mapping is established between the running state and the historical runtime time. Each historical runtime time corresponds to one running state. ,and ; The specific implementation process of step S3 includes: Based on the mapping relationship, the runtime state transition events are sorted out. If a runtime state transition occurs between two adjacent historical runtime nodes, it is marked as a runtime state transition event. The runtime state transition is represented as... ,in, And i, j ≠ v; identify all runtime state transition events and generate a runtime state transition matrix, denoted as: ,in, Indicates the running status Transition to running state ,Right now ; Based on the operational state transition matrix and the historical operational node time set, the number of each type of operational state transition event is counted, denoted as . Based on the number of each type of runtime state transition event, calculate the runtime state transition probability of any given event. The specific calculation formula is as follows: ,in, Indicates by The probability of state transition for a state transition event; Calculate running status Transition to running state The probability of state transition and the running state Transition to running state The error transition probability between the operating state transition probabilities is calculated using the following formula: ,in, Indicates the error transition probability. Indicates the preset operating state The perturbation weights, This indicates the preset error term. Indicates the running status to running status The probability of state transition for a state transition event.

2. The deep learning-based attitude stabilization control method for a marine rocket launch platform according to claim 1, characterized in that, The specific implementation process of step S1 includes: By employing sensor and data processing analysis technologies, historical operational status parameters of the sea-based rocket launch platform are acquired during its historical operations. These parameters include historical attitude angle data, historical roll rate data, and historical environmental disturbance force data. Based on these parameters, the operational status of the launch platform is determined, as follows: Preset attitude angle data range, roll rate data range, and environmental disturbance force data range; If the historical attitude angle data is less than or equal to the minimum value of the attitude angle data range, the historical roll rate data is less than or equal to the minimum value of the roll rate data range, and the historical environmental disturbance force data is less than or equal to the minimum value of the environmental disturbance force data range, then the operating state is determined to be a stable state. If the historical attitude angle data is within the attitude angle data range, the historical roll rate data is within the roll rate data range, and the historical environmental disturbance force data is within the environmental disturbance force data range, then the operating state is determined to be a moderate disturbance state. If the historical attitude angle data is greater than or equal to the maximum value of the attitude angle data range, the historical roll rate data is greater than or equal to the maximum value of the roll rate data range, and the historical environmental disturbance force data is greater than or equal to the maximum value of the environmental disturbance force data range, then the operating state is determined to be a strong disturbance state.

3. The deep learning-based attitude stabilization control method for a marine rocket launch platform according to claim 2, characterized in that, The specific implementation process of step S4 includes: If the current running status is Extract all running state transition probabilities The operating state with the highest probability The next operational state, as initially predicted, is as follows: ; According to the next running state and error transition probability Adjust the target attitude angle threshold as follows: ,in, Indicates the target attitude angle threshold. Indicates running status The corresponding physical limit value of the attitude angle, This represents the preset correction coefficient for the current attitude deviation. Indicates running status The corresponding attitude angle, Indicates the current actual attitude angle. This indicates the preset adjustment coefficient. Indicates running status The total error transition probability to all other non-self states; A preset error threshold is set if the total error transition probability is... If the error threshold is greater than or equal to the target attitude angle threshold, then the target attitude angle threshold is used. The control policy is executed as follows: ,in, Indicates the attitude angle adjustment step size. This indicates the preset maximum attitude adjustment step size. This represents the preset attitude deviation threshold. This indicates the preset basic posture adjustment step size; If the total error transfer probability If the value is less than the error threshold, then directly switch to the target attitude angle threshold. .

4. A deep learning-based attitude stabilization control system for a marine rocket launch platform, executing the deep learning-based attitude stabilization control method for a marine rocket launch platform as described in any one of claims 1-3, characterized in that, The system includes: a data acquisition and status judgment module, a mapping module, a matrix construction and error calculation module, and a threshold calculation and control module; The data acquisition and status judgment module acquires historical attitude angle data, historical roll rate data, and historical environmental disturbance force data of the sea-based rocket launch platform during its historical operation; and judges the operating status of the launch platform, which includes a stable state, a moderate disturbance state, and a strong disturbance state. The mapping module: counts all historical operating node times of the launch platform and generates a set of historical operating node times; maps the operating status to the historical operating node times. The matrix construction and error calculation module: marks the running state transition events according to the mapping relationship; identifies all running state transition events and generates a running state transition matrix; calculates the running state transition probability of any running state transition event; and calculates the error transition probability between the running state transition probabilities. The threshold calculation and control module extracts all operating state transition probabilities and uses the operating state with the highest probability as the next operating state for preliminary prediction; calculates the target attitude angle threshold; presets an error threshold; if the total error transition probability is greater than or equal to the error threshold, the control strategy is executed; if the total error transition probability is less than the error threshold, the system directly switches to the target attitude angle threshold.

5. The deep learning-based attitude stabilization control system for a marine rocket launch platform according to claim 4, characterized in that: The data acquisition and status determination module includes a data acquisition unit and a status determination unit; The data acquisition unit acquires historical operational status parameters of the sea-based rocket launch platform during its historical operation through sensors and data processing and analysis technology. These historical operational status parameters include historical attitude angle data, historical roll rate data, and historical environmental disturbance force data. Based on the historical attitude angle data, historical roll rate data, and historical environmental disturbance force data... The state determination unit determines the operating state of the launch platform as follows: It presets an attitude angle data range, a roll rate data range, and an environmental interference force data range. If the historical attitude angle data is less than or equal to the minimum value of the attitude angle data range, the historical roll rate data is less than or equal to the minimum value of the roll rate data range, and the historical environmental interference force data is less than or equal to the minimum value of the environmental interference force data range, then the operating state is determined to be a stable state. If the historical attitude angle data is within the attitude angle data range, the historical roll rate data is within the roll rate data range, and the historical environmental interference force data is within the environmental interference force data range, then the operating state is determined to be a moderate disturbance state. If the historical attitude angle data is greater than or equal to the maximum value of the attitude angle data range, the historical roll rate data is greater than or equal to the maximum value of the roll rate data range, and the historical environmental interference force data is greater than or equal to the maximum value of the environmental interference force data range, then the operating state is determined to be a strong disturbance state.

6. The deep learning-based attitude stabilization control system for a marine rocket launch platform according to claim 5, characterized in that: The mapping module includes a mapping unit; The mapping unit: counts all historical operating node times of the launch platform and generates a set of historical operating node times; based on the set of historical operating node times, maps the operating state to the historical operating node times in a time-to-operating state correspondence, wherein one historical operating node time corresponds to one operating state.

7. The deep learning-based attitude stabilization control system for a marine rocket launch platform according to claim 6, characterized in that: The matrix construction and error calculation module includes a matrix construction unit and an error calculation unit; The matrix construction unit: Based on the mapping relationship, sorts out the running state transition events. If a running state transition occurs between two adjacent historical running node times, it is marked as a running state transition event. All runtime state transition events are identified, and a runtime state transition matrix is ​​generated; The error calculation unit: counts the number of each type of operation state transition event based on the operation state transition matrix and the historical operation node time set; Calculate the state transition probability of any given state transition event based on the number of each type of state transition event. Calculate the error transition probability between the operating state transition probabilities.

8. The deep learning-based attitude stabilization control system for a marine rocket launch platform according to claim 7, characterized in that: The threshold calculation and control module includes a threshold calculation unit and a control unit; The threshold calculation unit extracts all the running state transition probabilities and takes the running state with the highest probability as the next running state for preliminary prediction. Adjust the target attitude angle threshold based on the next operating state and the error transfer probability; The control unit: presets an error threshold; if the total error transfer probability is greater than or equal to the error threshold, it executes a control strategy based on the target attitude angle threshold. If the total error transfer probability is less than the error threshold, then directly switch to the target attitude angle threshold.

Images