Rocket platform sea state adaptive lift control method based on self-supervised learning

Abstract

The application discloses a rocket platform sea state adaptive lifting control method based on self-supervised learning, S1. A pre-processed sea state disturbance data set is generated; S2. A multi-scale disturbance feature representation set is output; S3. A dynamic prior variational disturbance encoder is obtained; S4. The current pre-processed sea state disturbance data set is inferred by using the dynamic prior variational disturbance encoder, and a real-time sea state disturbance feature vector is output, and a control performance evaluation function is set; S5. The real-time sea state disturbance feature vector is used to guide the secretary bird population initialization position, and the optimal lifting control parameter group is obtained through iterative calculation of the dynamic guiding secretary bird optimization algorithm; S6. The optimal lifting control parameter group is input into the rocket platform lifting execution system and the platform lifting control is completed, and the platform lifting response data is generated. The application solves the problems that the traditional disturbance feature extraction method has insufficient recognition ability for non-structural strong noise signals and the disturbance prior hypothesis is too static, and greatly improves the disturbance perception accuracy and system robustness.

Description

Technical Field

[0001] This invention relates to the field of rocket platform technology, and in particular to a method for adaptive sea state ascent and descent control of rocket platforms based on self-supervised learning. Background Technology

[0002] As rocket platform applications expand to complex environments such as near-shore launches and sea-based recovery, the dynamic disturbances faced by the platforms during mission execution are becoming increasingly complex. During ascent and descent, operational stability and control accuracy have a decisive impact on the overall safety and success rate of the flight mission. Currently, rocket platforms in actual deployment mainly rely on traditional controller designs, such as fixed-gain PID controllers, fuzzy controllers, or feedback control systems based on preset rule bases. While existing methods offer a certain degree of controllability under static or weak disturbance conditions, they often exhibit response lag, unstable regulation, and insufficient control robustness under severe sea state disturbances, multi-scale dynamic interference, and strong nonlinear response conditions.

[0003] Existing rocket platform control systems generally face two key challenges: First, sea state disturbance signals are characterized by high dimensionality, multiple sources, and strong noise interference. Most of the disturbance sensing modules currently relied upon by control systems adopt simple filtering, threshold judgment, or empirical feature extraction methods based on the original sensor signals, which makes it difficult to extract high-quality disturbance features from complex signals, resulting in control input lag or even failure. Second, control parameters are multidimensional and complex and exhibit strong nonlinear dependencies. Traditional control strategies based on manual experience or linear parameter tuning cannot be adapted in real time in time-varying disturbance scenarios, and are prone to getting trapped in local optima or runaway states, failing to meet the rapid response requirements of rocket platform lifting systems in highly dynamic environments.

[0004] In addition, although some studies in recent years have begun to try to introduce intelligent optimization algorithms such as particle swarm optimization, genetic algorithms or reinforcement learning strategies to assist in the optimization of rocket platform control systems, the relevant methods have problems such as slow algorithm convergence speed, strong initialization sensitivity, inability to effectively integrate high-dimensional disturbance perception data, and most of them lack the ability to represent and model the disturbance data itself, making it difficult to form a dynamic adjustment mechanism for control parameters oriented towards disturbance structures.

[0005] Therefore, it is urgent to construct an integrated perception and control method for multi-scale sea state disturbances of rocket platforms to solve key problems such as inaccurate disturbance feature extraction, untimely update of control parameters, and unstable optimization strategies, so as to meet the stable ascent and descent control requirements of future rocket platforms in highly dynamic and complex environments. Summary of the Invention

[0006] One objective of this invention is to propose a self-supervised learning-based sea state adaptive ascent and descent control method for rocket platforms. This invention effectively solves the problems of insufficient ability of traditional disturbance feature extraction methods to identify unstructured strong noise signals and overly static prior assumptions about disturbances, and significantly improves the disturbance perception accuracy and system robustness.

[0007] An adaptive sea state ascent and descent control method for a rocket platform based on self-supervised learning, according to an embodiment of the present invention, includes the following steps:

[0008] S1. A data acquisition system is deployed synchronously on the rocket platform to continuously acquire and preprocess sea state disturbance datasets, generating preprocessed sea state disturbance datasets.

[0009] S2. Construct a multi-scale self-supervised variational contrastive learning model, and perform self-supervised training with a preprocessed sea state disturbance dataset as input, outputting a set of multi-scale disturbance feature representations;

[0010] S3. Based on the multi-scale perturbation feature representation set, construct a dynamic prior distribution model of sea state perturbation of rocket platform, and embed the dynamic prior distribution model into multi-scale self-supervised variational contrastive learning model to complete model update, and obtain dynamic prior variational perturbation encoder;

[0011] S4. Use a dynamic prior variational disturbance encoder to infer the current preprocessed sea state disturbance dataset, output the real-time sea state disturbance feature vector, and construct a multi-dimensional search space for lifting control parameters based on the real-time sea state disturbance feature vector and the multi-scale disturbance feature representation set, and set the control performance evaluation function.

[0012] S5. The real-time sea state disturbance feature vector is used to guide the initial position of the secretary bird population. The dynamic guided secretary bird optimization algorithm is started to perform a global search in the multi-dimensional search space of the lifting control parameters. The optimal lifting control parameter set is obtained by iterative calculation through the dynamic guided secretary bird optimization algorithm.

[0013] S6. Input the optimal lifting and lowering control parameter set into the rocket platform lifting and lowering execution system and complete the platform lifting and lowering control, generating platform lifting and lowering response data.

[0014] Optionally, S1 includes the following steps:

[0015] S11. By deploying accelerometers, gyroscopes, and sea state monitors at key structural locations on the rocket platform, and setting sampling time intervals and total sampling durations, disturbance signals of the rocket platform under dynamic sea state environments are continuously collected to form an original sea state disturbance dataset. Each disturbance sampling data point in the original sea state disturbance dataset is d. i Including timestamp t i , three-dimensional acceleration vector a i 3D angular velocity vector ωi and instantaneous effective wave height h i The total number of perturbation data is N raw =T total / Δt;

[0016] S12. Perform data preprocessing on the original sea state disturbance dataset, using a time window of length T. smooth The sliding filter method smooths the three-dimensional acceleration vector, three-dimensional angular velocity vector and instantaneous effective wave height value, eliminates high-frequency disturbance components caused by sensor sampling jitter, and performs anomaly detection and elimination of the signal based on the adaptive threshold mechanism, excludes atypical disturbance samples affected by external abnormal interference, and obtains the intermediate processing dataset after filtering and denoising.

[0017] S13. Perform perturbation feature enhancement processing on the intermediate processing dataset, calculate the perturbation rate of change feature between continuous perturbation data based on the time step, and extract the three-dimensional acceleration rate of change Δa. i 3D angular velocity change rate Δω i With the effective wave height change rate Δh i These represent the acceleration fluctuation trend, attitude disturbance fluctuation trend, and sea state disturbance fluctuation trend of the rocket platform at the current moment relative to the previous moment, respectively. The disturbance rate of change characteristics are concatenated with the corresponding three-dimensional acceleration vector, three-dimensional angular velocity vector, and instantaneous significant wave height value to form an enhanced disturbance data sample set. This enhanced disturbance data sample set is used as the final preprocessed sea state disturbance dataset D. pre .

[0018] Optionally, S2 includes the following steps:

[0019] S21. Construct a multi-scale self-supervised variational contrastive learning model to preprocess the sea state disturbance dataset D. pre As input, for the multi-frequency oscillation characteristics contained in the perturbation sample, a scale set S is defined, and each scale s in the scale set S... k The response window corresponding to different frequency disturbance characteristics in the rocket platform's ascent and descent system is obtained through a scale filter. Applying this to the channel splicing feature vectors to construct scale perturbation feature signals

[0020]

[0021] S22. Calculate the energy of the scale perturbation characteristic signal to measure its potential impact on the lift-down stability of the rocket platform. Define the scale energy as:

[0022]

[0023] in, Indicates the perturbation sample di At scale s k The total energy under the scale is expressed as the sum of squares of the combined signal, representing the driving amplitude of the disturbance signal on the dynamic response capability of the rocket platform at this scale.

[0024] S23. Normalize the scale energy to obtain the scale energy weights of the perturbed samples.

[0025]

[0026] in, This indicates that the perturbation sample is at scale s k The scale energy weight reflects the importance of perturbations at that scale to the rocket platform's ascent and descent decisions.

[0027] S24. For the perturbation characteristic signal at each scale Set up and input the global disturbance encoder in parallel. With local disturbance encoder Extract the slowly varying trend disturbances and instantaneous disturbances during the rocket platform's ascent and descent:

[0028]

[0029] in, Representing scale s k The mean vector of the lower perturbation characteristic distribution. For scale s k The standard deviation vector of the lower perturbation characteristic distribution;

[0030] S25. Generate standard normal noise independently for each scale based on a reparameterization strategy. and By combining the corresponding mean vector and standard deviation vector, the long-term perturbation latent vector is obtained. With the latent vector of instantaneous perturbation

[0031]

[0032] S26. Calculate the long-term perturbation latent vector. With the latent vector of instantaneous perturbation The cosine similarity between them, combined with the latent variable dimension d z Temperature scaling is performed, and then the softmax function is applied to obtain the cross-scale attention coefficients. Cross-scale attention coefficients reflect the mutual driving relationship of perturbations at different scales and are used to construct coordinated rise and fall control signal pathways;

[0033] S27. According to scale energy weight and cross-scale attention coefficient For instantaneous disturbance latent vector Perform a weighted summation to generate the perturbation fusion vector f. i The disturbance fusion vector integrates the combined effects of sea state disturbances at various scales on the rocket platform's ascent and descent at the current moment:

[0034]

[0035] S28. Fusion perturbation vector f i Input disturbance stability prediction head P θ Output the predicted slope value of future wave height change. And based on the actual wave height variation trend of the rocket platform under the current disturbance, s i Compared with the predicted value The mean square error between them defines the stability prediction loss L. p This is used to measure the accuracy of predictions regarding the future stable trend of a rocket platform.

[0036]

[0037] Where, N pre This represents the total number of perturbation samples used to train the perturbation stability prediction head;

[0038] S29. By jointly optimizing the multi-scale variational reconstruction loss L v Self-supervised contrast loss L c With stability prediction loss L p The parameters of the multi-scale self-supervised variational contrastive learning model are updated by adjusting the relative weights of various losses in the sea state disturbance perception task using adjustment coefficients λ1 and λ2:

[0039] L total =L v +λ1L c +λ2L p ;

[0040] Wherein, λ1 and λ2 are the adjustment coefficients for contrastive learning and prediction loss, respectively, controlling the relative weights of feature learning and stability judgment in the perturbation perception task;

[0041] S210. After completing the joint training, output a set of multi-scale perturbation feature representations.

[0042] Optionally, S3 includes the following steps:

[0043] S31. Based on the multi-scale perturbation feature representation set F multi According to the timestamp t of the perturbation sample i Construct the temporal trajectory of the perturbation features, and fuse each perturbation vector f i Its corresponding timestamp ti The set of temporal state trajectories T, which constitutes the disturbance characteristics, is used to characterize the evolution of sea state disturbances of the rocket platform over a continuous time period.

[0044] S32. Set a sliding time window T of fixed length on the perturbation characteristic temporal state trajectory set T. prior The time series trajectory set of disturbance features is divided into several disturbance local sub-window sequences by using a sliding time window. Each disturbance local sub-window sequence contains a time period [t]. j ,t j +T prior The set of all fused perturbation feature vectors within a given region is denoted as the perturbation feature sub-window set W. j It is used to extract the local disturbance distribution pattern of the rocket platform at different stages of sea state change;

[0045] S33. For each set of perturbation feature sub-windows W j The fused perturbation feature vectors are statistically analyzed, and the mean vector of all perturbation features in the perturbation feature sub-window set is calculated. With covariance matrix These are used to describe the trend of the disturbance center and the spread range of the disturbance characteristics of the rocket platform in the current time period, respectively.

[0046] S34. Constructing a priori prediction network P φ The timestamp t of the perturbation sample i and the corresponding perturbation fusion vector f i Input it, output the estimated prior distribution parameters To estimate the potential response structure of the rocket platform to sea state disturbance at the current moment, and to define the prior estimation error loss function using KL divergence:

[0047]

[0048] in, These are the posterior distribution parameters calculated by the multi-scale encoder at the current time. The KL divergence measures the structural deviation between the current state of the platform and the predicted perturbation trend, which is the estimated prior distribution parameter of the prior prediction network output.

[0049] S35. Estimate the prior distribution parameters from the output of the prior prediction network. Embedded in a multi-scale self-supervised variational contrastive learning model, it replaces the standard normal distribution as a new variational prior distribution, completing the dynamic structural update of the disturbance perception model during the rocket platform's ascent and descent, thus forming a dynamic prior variational disturbance encoder.

[0050] Optionally, S4 includes the following steps:

[0051] S41. Based on the dynamic prior variational disturbance encoder, receive the latest enhanced disturbance data sample d from the current preprocessed sea state disturbance dataset. t And with its channel splicing feature x t Using this as input, infer the posterior distribution parameters of the perturbation at the current time step. Simultaneously, the prior prediction network generates the current prior distribution parameters. Based on the dynamic prior distribution, a reparameterization strategy is used to generate the real-time perturbation latent feature vector at the current moment:

[0052]

[0053] Among them, z t Let be the real-time potential perturbation feature vector of the platform in the multi-scale perturbation space at the current moment, ∈ t This is a standard normal noise vector, representing an abstract representation of the current sea state disturbance during the rocket platform's ascent and descent.

[0054] S42. Perturb the latent feature vector z in real time. t With the multi-scale perturbation feature representation set F multi Scale alignment is performed, and attention weights are obtained through a cosine similarity-softmax mechanism adjusted by a temperature coefficient τ. Then apply attention weights to the perturbation fusion vector f i Weighted summation yields the perturbation-driven fusion vector.

[0055] S43. Based on the perturbation-driven fusion vector Construct a multidimensional search space P for lifting control parameters. The multidimensional search space for lifting control parameters is composed of the control parameter vector p = (p1, p2, ..., p...). d The control parameter vector consists of each control parameter p. i All are constrained to the corresponding minimum value With the maximum value between;

[0056] S44. Define a control performance evaluation function J(p,z) for the multidimensional search space P of the lifting control parameters. t The control performance evaluation function is based on the highly fluctuating residual ΔH. res (p,z t ), attitude stability error Δθ stab (p,z t and control response delay time T delay (p,z t The three weighted sums of squares, with weighting coefficients λ1, λ2, and λ3 respectively, are used to comprehensively measure the lift control accuracy and response efficiency of the rocket platform under the current potential disturbance state.

[0057] Optionally, S5 includes the following steps:

[0058] S51. Based on the real-time perturbation potential feature vector z t Calculate the current disturbance intensity index I t Among them, the disturbance intensity index I t Based on the real-time perturbation of the latent eigenvector z t The L2 norm is determined and used to measure the overall energy amplitude of the disturbance experienced by the rocket platform under current sea conditions;

[0059] S52. Construct a population of secretary birds containing M individuals within the multidimensional search space P of the elevation control parameters, and perturb the latent feature vector z in real time. t As heuristic information, through the weight matrix W (m) With bias vector b (m) After linear transformation and compression using the Sigmoid function, the perturbation mapping coefficients Γ are obtained. (m) Then, according to the lower bound vector p min With upper limit vector p max The interval proportional mapping method generates the initial control parameter vector for each secretary bird. Complete the perturbation-driven initialization of the secretary bird population;

[0060] S53. Based on disturbance intensity index I t Set the adaptive step size adjustment factor α t Adaptive step size adjustment factor α t Based on the basic step size α base Together with the step size decay coefficient κ, the disturbance intensity index I is determined when... t When the step size is increased, the adaptive step size adjustment factor α t Decrease according to the inverse proportional relationship;

[0061] S54. During the secretary bird optimization iteration process, for each secretary bird in the kth generation, the current control parameter vector is updated using a jump-tracking update strategy. Add adaptive step size adjustment factor α t The directional increment, mixed with random weights r1 and r2, achieves the goal of moving towards the optimal control parameter vector. With random individual vectors Simultaneously converge, thereby obtaining the next generation control parameter vector.

[0062]

[0063] in, J(p,z) represents the control performance evaluation function in the kth generation. t The optimal control parameter vector. r1 and r2 are random individual vectors in the current population, and are random weights that are independently and uniformly sampled in the interval [0,1], used to simulate the hunting focus behavior and global reconnaissance behavior of the secretary bird.

[0064] S55. Repeat the secretary bird population update until the set termination condition is reached. The termination condition includes a maximum iteration limit or a control performance evaluation function J(p,z) for a certain number of consecutive generations. t If the improvement is less than the threshold, the final output will satisfy the control performance evaluation function J(p,z) throughout the entire iteration process. t The control parameter vector p that achieves the minimum value * Control parameter vector p * This serves as the current optimal set of lifting control parameters for the rocket platform's lifting and lowering execution system.

[0065] Optionally, S6 includes the following steps:

[0066] S61. Transfer the optimal lift control parameter vector p * Input the rocket platform's ascent and descent execution system to drive the platform to execute the current ascent and descent task. The ascent and descent execution system adjusts the ascent and descent rate, attitude angle stability factor, and disturbance compensation gain key control signals according to the strategy values ​​corresponding to each control dimension in the optimal control parameter vector.

[0067] S62. During the ascent and descent process, a sensor system fixedly connected to the rocket platform structure monitors the dynamic process in real time, recording the platform's altitude change curve, attitude change trajectory, attitude deviation rate, and ascent and descent time delay platform response variables within the ascent and descent response cycle, generating a platform ascent and descent response dataset R. exec ;

[0068] S63. Platform Lifting / Rising Response Dataset R exec Each response data r i Includes the following parameter field: the platform's current elevation height H. i Platform attitude angle change rate Δθ i Control command response time Disturbance compensation error System control status label L i System control status label L i The description uses labels to indicate the current operating status of the platform.

[0069] Optionally, the label identifier is defined as follows:

[0070] Normal operating condition: Disturbance compensation error Response time Attitude angle change rate Δθ i ≤3° / s;

[0071] Disturbed operating status: Disturbance compensation error Response time Or the rate of change of attitude angle is 3° / s < Δθ i ≤5° / s;

[0072] Control instability: Disturbance compensation error Response time Or the rate of change of attitude angle Δθ i >5° / s.

[0073] The beneficial effects of this invention are:

[0074] (1) This invention proposes a multi-scale variational contrastive learning architecture in the disturbance perception stage. By introducing scale energy distribution and a self-attention cross-scale collaborative mechanism, the model can extract the temporal variation patterns of multimodal information such as acceleration, angular velocity and significant wave height from disturbance response windows of different frequencies, and obtain a stable representation by integrating the variational sampling mechanism. On this basis, a dynamic disturbance prior distribution network constructed by embedding time series is used to replace the traditional fixed standard normal distribution. By predictively modeling the evolution trend of disturbance distribution, the model's sensitivity and stability to the temporal features of sea state disturbances are enhanced. This effectively solves the problems of insufficient ability to identify unstructured strong noise signals and overly static disturbance prior assumptions in traditional disturbance feature extraction methods, and significantly improves the disturbance perception accuracy and system robustness.

[0075] (2) This invention constructs a secretary bird population initialization strategy guided by the potential feature vector of disturbance. Combined with the dynamic generation of the initialization control parameter position distribution by the disturbance intensity, the secretary bird population starts from the highly correlated region of the disturbance state, skipping the inefficient exploration process caused by a large number of random initializations in the traditional algorithm. At the same time, the step size adjustment coefficient is driven by the disturbance intensity index to construct a two-way closed-loop mechanism of disturbance perception-optimization search. When the sea state changes drastically, it dynamically converges to a high-precision local region and maintains global search capability under stable disturbance conditions. The improved strategy is significantly better than the traditional secretary bird and particle swarm fixed step size optimization algorithms. In high-dimensional parameter space and nonlinear control function scenarios, the speed of control parameter optimization is improved.

[0076] (3) This invention establishes a direct feedback channel between the disturbance perception model and the platform lifting and lowering execution system. The attitude change rate, wave height response, and control delay indicators generated in real time during the lifting and lowering execution process constitute the platform lifting and lowering response dataset, and the operating status labels are defined according to the disturbance compensation error and attitude stability. This is not only used to evaluate the effect of the current control strategy, but also to apply the structured feedback information in reverse to the dynamic prior prediction network and the secretary bird optimizer, for online updating of the control strategy and the disturbance prior model, realizing intelligent adaptive control under disturbance-driven conditions. This linked closed-loop system is significantly superior to the traditional "perception-control-execution" unidirectional architecture, and has extremely high dynamic adaptability and engineering practicality. Attached Figure Description

[0077] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0078] Figure 1 This is a flowchart of a self-supervised learning-based adaptive sea state lift control method for rocket platforms proposed in this invention. Detailed Implementation

[0079] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0080] refer to Figure 1 A self-supervised learning-based adaptive sea state lift control method for rocket platforms includes the following steps:

[0081] S1. A data acquisition system is deployed synchronously on the rocket platform to continuously acquire and preprocess sea state disturbance datasets, generating preprocessed sea state disturbance datasets.

[0082] S2. Construct a multi-scale self-supervised variational contrastive learning model, and perform self-supervised training with a preprocessed sea state disturbance dataset as input, outputting a set of multi-scale disturbance feature representations;

[0083] S3. Based on the multi-scale perturbation feature representation set, construct a dynamic prior distribution model of sea state perturbation of rocket platform, and embed the dynamic prior distribution model into multi-scale self-supervised variational contrastive learning model to complete model update, and obtain dynamic prior variational perturbation encoder;

[0084] S4. Use a dynamic prior variational disturbance encoder to infer the current preprocessed sea state disturbance dataset, output the real-time sea state disturbance feature vector, and construct a multi-dimensional search space for lifting control parameters based on the real-time sea state disturbance feature vector and the multi-scale disturbance feature representation set, and set the control performance evaluation function.

[0085] S5. The real-time sea state disturbance feature vector is used to guide the initial position of the secretary bird population. The dynamic guided secretary bird optimization algorithm is started to perform a global search in the multi-dimensional search space of the lifting control parameters. The optimal lifting control parameter set is obtained by iterative calculation through the dynamic guided secretary bird optimization algorithm.

[0086] S6. Input the optimal lifting and lowering control parameter set into the rocket platform lifting and lowering execution system and complete the platform lifting and lowering control, generating platform lifting and lowering response data.

[0087] In this embodiment, S1 includes the following steps:

[0088] S11. By deploying accelerometers, gyroscopes, and sea state monitors at key structural locations on the rocket platform, and setting sampling time intervals and total sampling durations, disturbance signals of the rocket platform under dynamic sea state environments are continuously collected to form an original sea state disturbance dataset. Each disturbance sampling data point in the original sea state disturbance dataset is d. i Including timestamp t i , three-dimensional acceleration vector a i 3D angular velocity vector ω i and instantaneous effective wave height h i The total number of perturbation data is N raw =T total / Δt;

[0089] S12. Perform data preprocessing on the original sea state disturbance dataset, using a time window of length T. smooth The sliding filter method smooths the three-dimensional acceleration vector, three-dimensional angular velocity vector and instantaneous effective wave height value, eliminates high-frequency disturbance components caused by sensor sampling jitter, and performs anomaly detection and elimination of the signal based on the adaptive threshold mechanism, excludes atypical disturbance samples affected by external abnormal interference, and obtains the intermediate processing dataset after filtering and denoising.

[0090] S13. Perform perturbation feature enhancement processing on the intermediate processing dataset, calculate the perturbation rate of change feature between continuous perturbation data based on the time step, and extract the three-dimensional acceleration rate of change Δa. i 3D angular velocity change rate Δω i With the effective wave height change rate Δh i These represent the acceleration fluctuation trend, attitude disturbance fluctuation trend, and sea state disturbance fluctuation trend of the rocket platform at the current moment relative to the previous moment, respectively. The disturbance rate of change characteristics are concatenated with the corresponding three-dimensional acceleration vector, three-dimensional angular velocity vector, and instantaneous significant wave height value to form an enhanced disturbance data sample set. This enhanced disturbance data sample set is used as the final preprocessed sea state disturbance dataset D. pre .

[0091] In this embodiment, S2 includes the following steps:

[0092] S21. Construct a multi-scale self-supervised variational contrastive learning model to preprocess the sea state disturbance dataset D. pre As input, for the multi-frequency oscillation characteristics contained in the perturbation sample, a scale set S is defined, and each scale s in the scale set S... k The response window corresponding to different frequency disturbance characteristics in the rocket platform's ascent and descent system is obtained through a scale filter. Applying this to the channel splicing feature vectors to construct scale perturbation feature signals

[0093]

[0094] S22. Calculate the energy of the scale perturbation characteristic signal to measure its potential impact on the lift-down stability of the rocket platform. Define the scale energy as:

[0095]

[0096] in, Indicates the perturbation sample d i At scale s k The total energy under the scale is expressed as the sum of squares of the combined signal, representing the driving amplitude of the disturbance signal on the dynamic response capability of the rocket platform at this scale.

[0097] S23. Normalize the scale energy to obtain the scale energy weights of the perturbed samples.

[0098]

[0099] in, This indicates that the perturbation sample is at scale s k The scale energy weight reflects the importance of perturbations at that scale to the rocket platform's ascent and descent decisions.

[0100] S24. For the perturbation characteristic signal at each scale Set up and input the global disturbance encoder in parallel. With local disturbance encoder Extract the slowly varying trend disturbances and instantaneous disturbances during the rocket platform's ascent and descent:

[0101]

[0102] in, Representing scale s k The mean vector of the lower perturbation characteristic distribution. For scale s k The standard deviation vector of the lower perturbation characteristic distribution;

[0103] S25. Generate standard normal noise independently for each scale based on a reparameterization strategy. and By combining the corresponding mean vector and standard deviation vector, the long-term perturbation latent vector is obtained. With the latent vector of instantaneous perturbation

[0104]

[0105]

[0106] S26. Calculate the long-term perturbation latent vector. With the latent vector of instantaneous perturbation The cosine similarity between them, combined with the latent variable dimension d z Temperature scaling is performed, and then the softmax function is applied to obtain the cross-scale attention coefficients. Cross-scale attention coefficients reflect the mutual driving relationship of perturbations at different scales and are used to construct coordinated rise and fall control signal pathways;

[0107] S27. According to scale energy weight and cross-scale attention coefficient For instantaneous disturbance latent vector Perform a weighted summation to generate the perturbation fusion vector f. i The disturbance fusion vector integrates the combined effects of sea state disturbances at various scales on the rocket platform's ascent and descent at the current moment:

[0108]

[0109] S28. Fusion perturbation vector f i Input disturbance stability prediction head P θ Output the predicted slope value of future wave height change. And based on the actual wave height variation trend of the rocket platform under the current disturbance, s i Compared with the predicted value The mean square error between them defines the stability prediction loss L. p This is used to measure the accuracy of predictions regarding the future stable trend of a rocket platform.

[0110]

[0111] Where, N pre This represents the total number of perturbation samples used to train the perturbation stability prediction head;

[0112] S29. By jointly optimizing the multi-scale variational reconstruction loss L v Self-supervised contrast loss L c With stability prediction loss Lp The parameters of the multi-scale self-supervised variational contrastive learning model are updated by adjusting the relative weights of various losses in the sea state disturbance perception task using adjustment coefficients λ1 and λ2:

[0113] L total =L v +λ1L c +λ2L p ;

[0114] Wherein, λ1 and λ2 are the adjustment coefficients for contrastive learning and prediction loss, respectively, controlling the relative weights of feature learning and stability judgment in the perturbation perception task;

[0115] Multiscale variational reconstruction loss L v This is used to constrain the generation capability of latent perturbation variables, enabling the model to accurately reconstruct perturbation features at various scales and maintain distributional consistency and structural fidelity between the latent perturbation representation and the observed data. Its construction logic is as follows:

[0116] For each scale s k Construct global perturbation encoders respectively With local disturbance encoder Extract the long-term mean vector of the perturbation sample at the current scale. Standard deviation vector and transient disturbances

[0117] A reparameterization strategy is employed to generate latent perturbation variables based on the aforementioned mean and standard deviation. and

[0118] Input latent variables into the perturbation reconstructor Reconstructed values ​​of the scale-perturbed signal

[0119] Compare real-scale perturbation signals With reconstructed values The difference is defined as follows: the reconstruction error loss at each scale is: observation error term (MSE loss) and KL divergence term: measures the deviation between the posterior and the prior (or standard normal). The sum of the losses at all scales constitutes the overall variational reconstruction loss. Used to maintain the integrity and multi-scale fidelity of perturbation information during the encoding-decoding process.

[0120] The multi-scale variational reconstruction loss term ensures that the model can effectively reconstruct disturbance characteristics at different scales (low-frequency wave trends and high-frequency jitter), maintain the rocket platform's ability to identify and recover multi-scale disturbances in complex sea conditions, and effectively support the accurate perception of subsequent lift-down control tasks.

[0121] Self-supervised contrastive loss is used to improve the discriminativeness and temporal structure consistency of perturbation features in the latent space. It achieves representation learning driven by the similarity between perturbations through a strategy of bringing positive samples closer and pushing negative samples further apart. The construction logic is as follows:

[0122] The perturbation sample d i Generate two positive sample representations under different data augmentation methods (scale perturbation, feature perturbation). From the same disturbance source;

[0123] Negative samples are drawn from other perturbation samples (at different times or under different perturbation scenarios) in the same batch.

[0124] The objective of contrastive learning is defined as: to make positive sample pairs... To maintain proximity in the feature space and distinguish between negative samples, a contrastive loss is constructed using the temperature-scaled InfoNCE loss form.

[0125]

[0126] Where sim(·) represents the cosine similarity function, and τ is the temperature adjustment coefficient.

[0127] In rocket platform ascent and descent missions, disturbances in different sea states may have similar trends but significant differences in detail. By using self-supervised contrastive loss, the model is able to identify the "fine-grained differences" of disturbances, thereby improving the model's response sensitivity to critical moments of sudden disturbance changes and platform attitude transitions, and thus supporting highly robust decision-making in ascent and descent control.

[0128] S210. After completing the joint training, output a set of multi-scale perturbation feature representations.

[0129] In this embodiment, S3 includes the following steps:

[0130] S31. Based on the multi-scale perturbation feature representation set F multi According to the timestamp t of the perturbation sample i Construct the temporal trajectory of the perturbation features, and fuse each perturbation vector f i Its corresponding timestamp t i The set of temporal state trajectories T, which constitutes the disturbance characteristics, is used to characterize the evolution of sea state disturbances of the rocket platform over a continuous time period.

[0131] S32. Set a sliding time window T of fixed length on the perturbation characteristic temporal state trajectory set T. prior The time series trajectory set of disturbance features is divided into several disturbance local sub-window sequences by using a sliding time window. Each disturbance local sub-window sequence contains a time period [t]. j ,tj +T prior The set of all fused perturbation feature vectors within a given region is denoted as the perturbation feature sub-window set W. j It is used to extract the local disturbance distribution pattern of the rocket platform at different stages of sea state change;

[0132] S33. For each set of perturbation feature sub-windows W j The fused perturbation feature vectors are statistically analyzed, and the mean vector of all perturbation features in the perturbation feature sub-window set is calculated. With covariance matrix These are used to describe the trend of the disturbance center and the spread range of the disturbance characteristics of the rocket platform in the current time period, respectively.

[0133] S34. Constructing a priori prediction network P φ The timestamp t of the perturbation sample i and the corresponding perturbation fusion vector f i Input it, output the estimated prior distribution parameters To estimate the potential response structure of the rocket platform to sea state disturbance at the current moment, and to define the prior estimation error loss function using KL divergence:

[0134]

[0135] in, These are the posterior distribution parameters calculated by the multi-scale encoder at the current time. The KL divergence measures the structural deviation between the current state of the platform and the predicted perturbation trend, which is the estimated prior distribution parameter of the prior prediction network output.

[0136] S35. Estimate the prior distribution parameters from the output of the prior prediction network. Embedded in a multi-scale self-supervised variational contrastive learning model, it replaces the standard normal distribution as a new variational prior distribution, completing the dynamic structural update of the disturbance perception model during the rocket platform's ascent and descent, thus forming a dynamic prior variational disturbance encoder.

[0137] In this embodiment, S4 includes the following steps:

[0138] S41. Based on the dynamic prior variational disturbance encoder, receive the latest enhanced disturbance data sample d from the current preprocessed sea state disturbance dataset. t And with its channel splicing feature x t Using this as input, infer the posterior distribution parameters of the perturbation at the current time step. Simultaneously, the prior prediction network generates the current prior distribution parameters. Based on the dynamic prior distribution, a reparameterization strategy is used to generate the real-time perturbation latent feature vector at the current moment:

[0139]

[0140] Among them, z t Let be the real-time potential perturbation feature vector of the platform in the multi-scale perturbation space at the current moment, ∈ t This is a standard normal noise vector, representing an abstract representation of the current sea state disturbance during the rocket platform's ascent and descent.

[0141] S42. Perturb the latent feature vector z in real time. t With the multi-scale perturbation feature representation set F multi Scale alignment is performed, and attention weights are obtained through a cosine similarity-softmax mechanism adjusted by a temperature coefficient τ. Then apply attention weights to the perturbation fusion vector f i Weighted summation yields the perturbation-driven fusion vector.

[0142] S43. Based on the perturbation-driven fusion vector Construct a multidimensional search space P for lifting control parameters. The multidimensional search space for lifting control parameters is composed of the control parameter vector p = (p1, p2, ..., p...). d The control parameter vector consists of each control parameter p. i All are constrained to the corresponding minimum value With the maximum value between;

[0143] S44. Define a control performance evaluation function J(p,z) for the multidimensional search space P of the lifting control parameters. t The control performance evaluation function is based on the highly fluctuating residual ΔH. res (p,z t ), attitude stability error Δθ stab (p,z t and control response delay time T delay (p,z t The three weighted sums of squares, with weighting coefficients λ1, λ2, and λ3 respectively, are used to comprehensively measure the lift control accuracy and response efficiency of the rocket platform under the current potential disturbance state.

[0144] In this embodiment, the height fluctuation residual ΔH res (p,z t ) is used to measure the potential state z under the current disturbance. t Under the condition of control parameter p, the degree of deviation of the rocket platform's ascent and descent altitude from the desired trajectory is constructed as follows:

[0145] During the lifting control process, the platform executes a lifting trajectory according to the control parameter p. Let the actual height sequence within this time period be... The unit is meters (m);

[0146] Let the expected height trajectory corresponding to this time period be... It can be generated by the target attitude planner;

[0147] The square root of the average residual between the actual trajectory and the expected trajectory over the entire response period is taken as the height fluctuation residual: the smaller the residual, the more stable the platform height change and the higher the control accuracy; the unit is still meters, consistent with the dimension of the height output by the lifting system.

[0148] Attitude stability error Δθ stab (p,z t This is used to characterize the dynamic changes in attitude angles of a rocket platform during ascent and descent, reflecting the degree of attitude control jitter during the process. The construction method is as follows:

[0149] During control execution, the changes in platform attitude angles over time are recorded to obtain a sequence of attitude angle change rates: The unit is degrees per second (° / s); it can be obtained from the platform attitude inertial measurement unit.

[0150] The variance or maximum rate of change of attitude within the response period is calculated as a stability index of the platform under that parameter condition; the smaller the variance / maximum value, the less platform jitter and the smoother the ascent and descent; finally, it is normalized to a single stability error index Δθ. stab (p,z t (), the unit is degrees per second.

[0151] Control response delay time T delay (p,z t This measure is used to assess the latency between receiving a control command and generating an effective action, reflecting the system's agile response capability under disturbances. It is constructed as follows:

[0152] For each control cycle, record the control command issuance time t. cmd The time t when the platform detects the first response action res ;

[0153] The control response delay time is defined as: T delay =t res -t cmd The platform's initial response can be determined by the abrupt change trend collected from the speed sensor or altitude sensor; the shorter the delay, the faster the system response; the unit is seconds (s).

[0154] In this embodiment, S5 includes the following steps:

[0155] S51. Based on the real-time perturbation potential feature vector z t Calculate the current disturbance intensity index It Among them, the disturbance intensity index I t Based on the real-time perturbation of the latent eigenvector z t The L2 norm is determined and used to measure the overall energy amplitude of the disturbance experienced by the rocket platform under current sea conditions;

[0156] S52. Construct a population of secretary birds containing M individuals within the multidimensional search space P of the elevation control parameters, and perturb the latent feature vector z in real time. t As heuristic information, through the weight matrix W (m) With bias vector b (m) After linear transformation and compression using the Sigmoid function, the perturbation mapping coefficients Γ are obtained. (m) Then, according to the lower bound vector p min With upper limit vector p max The interval proportional mapping method generates the initial control parameter vector for each secretary bird. Complete the perturbation-driven initialization of the secretary bird population;

[0157] S53. Based on disturbance intensity index I t Set the adaptive step size adjustment factor α t Adaptive step size adjustment factor α t Based on the basic step size α base Together with the step size decay coefficient κ, the disturbance intensity index I is determined when... t When the step size is increased, the adaptive step size adjustment factor α t Decrease according to the inverse proportional relationship;

[0158] S54. During the secretary bird optimization iteration process, for each secretary bird in the kth generation, the current control parameter vector is updated using a jump-tracking update strategy. Add adaptive step size adjustment factor α t The directional increment, mixed with random weights r1 and r2, achieves the goal of moving towards the optimal control parameter vector. With random individual vectors Simultaneously converge, thereby obtaining the next generation control parameter vector.

[0159]

[0160] in, J(p,z) represents the control performance evaluation function in the kth generation. t The optimal control parameter vector. r1 and r2 are random individual vectors in the current population, and are random weights that are independently and uniformly sampled in the interval [0,1], used to simulate the hunting focus behavior and global reconnaissance behavior of the secretary bird.

[0161] S55. Repeat the secretary bird population update until the set termination condition is reached. The termination condition includes a maximum iteration limit or a control performance evaluation function J(p,z) for a certain number of consecutive generations. t If the improvement is less than the threshold, the final output will satisfy the control performance evaluation function J(p,z) throughout the entire iteration process. t The control parameter vector p that achieves the minimum value * Control parameter vector p * This serves as the current optimal set of lifting control parameters for the rocket platform's lifting and lowering execution system.

[0162] In this embodiment, S6 includes the following steps:

[0163] S61. Transfer the optimal lift control parameter vector p * Input the rocket platform's ascent and descent execution system to drive the platform to execute the current ascent and descent task. The ascent and descent execution system adjusts the ascent and descent rate, attitude angle stability factor, and disturbance compensation gain key control signals according to the strategy values ​​corresponding to each control dimension in the optimal control parameter vector.

[0164] S62. During the ascent and descent process, a sensor system fixedly connected to the rocket platform structure monitors the dynamic process in real time, recording the platform's altitude change curve, attitude change trajectory, attitude deviation rate, and ascent and descent time delay platform response variables within the ascent and descent response cycle, generating a platform ascent and descent response dataset R. exec ;

[0165] S63. Platform Lifting / Rising Response Dataset R exec Each response data r i Includes the following parameter field: the platform's current elevation height H. i Platform attitude angle change rate Δθ i Control command response time Disturbance compensation error System control status label L i The meanings of each parameter are as follows:

[0166] Lifting height H i The actual height value of the platform at the i-th sampling moment after executing the control command, in meters (m);

[0167] Attitude angle change rate Δθ i The change in attitude angle per unit time during the platform's ascent and descent reflects the platform's stability, and is expressed in degrees per second (° / s).

[0168] Control command response time The time delay from the issuance of a control command to its completion of execution on the platform, measured in seconds (s);

[0169] Disturbance compensation error The deviation between the platform's expected attitude state and the actual attitude state, expressed in units consistent with attitude angles, represents the accuracy of disturbance compensation;

[0170] System control status label L i : Label the current operating status of the platform.

[0171] In this embodiment, the label identifier is defined as follows:

[0172] Normal operating condition: Disturbance compensation error Response time Attitude angle change rate Δθ i ≤3° / s;

[0173] Disturbed operating status: Disturbance compensation error Response time Or the rate of change of attitude angle is 3° / s < Δθ i ≤5° / s;

[0174] Control instability: Disturbance compensation error Response time Or the rate of change of attitude angle Δθ i >5° / s.

[0175] Example 1:

[0176] This invention describes the complete perception-optimization-control process of a rocket platform encountering complex, multi-scale sea state disturbances before ascent and descent. During a pre-launch inspection of a rocket platform at sea, the system planned to execute a standard ascent and descent simulation to test the platform's responsiveness. The platform was equipped with this invention. During the ascent and descent preparation phase, the platform was in a high-frequency disturbance band, and the system continuously recorded multiple disturbance surges.

[0177] During the T-60-second lift preparation period, the platform's sensor system is activated, with three types of sensors deployed at the bottom and four corners of the platform beginning data sampling at a frequency of 50Hz:

[0178] Accelerometer data returned: timestamp t 2218 = 60.3 seconds, three-dimensional acceleration vector a 2218 = (0.42, -0.18, 1.54) m / s 2 ; Gyroscope returned data: angular velocity vector ω 2218 = (0.05, 0.11, -0.07) rad / s; Wave height monitored by sea state instrument: significant wave height h 2218 =2.37m.

[0179] Data constitutes perturbation sample d 2218 This data is then merged with historical samples to form the original sea state disturbance dataset. Then it undergoes sliding window filtering (window size T) smooth =0.6s), and use the rate of change of acceleration Δa i Angular velocity change rate Δω i Wave height change rate Δh i Construct perturbation-enhanced samples to generate a preprocessed dataset D. pre .

[0180] The system will preprocess dataset D pre The input is fed into a multi-scale self-supervised variational contrastive learning model, which is configured with three scales S = {s1 = 0.5s, s2 = 1.0s, s3 = 2.0s}. At each scale:

[0181] For concatenated feature vectors x i =a i ||ω i ||h i ||Δa i ||Δω i ||Δh i Application scale filter Extracting scale perturbation features Calculate its scale energy For example, the maximum energy of 7.94 is obtained at the high-frequency scale s3; the scale attention weight is obtained. To reflect the importance of this scale of disturbance to lift control; extract long-term / short-term disturbance latent vectors. Calculate the cross-scale attention coefficients, and generate a perturbation feature vector f after fusion. i The platform outputs the current disturbance characteristic representation at the current time t=60.3s.

[0182] The system initiates the dynamic disturbance modeling module at t=60.3s:

[0183] Sliding window T prior =5s, covering the past 250 samples; extract the perturbation vector set W within the window. j ={f 1968 ,...,f 2218} Calculate the prior mean With covariance Prior prediction network P φ Receive f 2218 ,t 2218 Output estimate prior distribution This distribution is used as the basis for the next variational inference step to complete the dynamic prior variational perturbation encoder update.

[0184] The system will f 2218 and Combined with reparameterization to generate the perturbation potential vector z2218 ,2-norm ||z 2218 ||2 = 1.98. (z) 2218 Based on:

[0185] Construct a control parameter vector p = (p1: acceleration / deceleration rate, p2: stability coefficient, p3: disturbance compensation gain) ∈ P; define the evaluation function:

[0186]

[0187] The control objective is to minimize platform lifting offset and delay time.

[0188] According to the disturbance intensity I 2218 =1.98, the system adaptively sets the step size adjustment factor α 2218 =0.014, initialize the secretary bird population M=20: the position of each secretary bird is determined by the perturbation vector z. 2218 Projected and control parameter upper and lower limits p min ,p max Interval mapping generation; 20 rounds of secretary bird jump-tracking strategy iterations are performed, with the control vector updated in each round. The optimal parameter vector p is obtained in the 16th round. * = (Rise / fall rate = 1.07, stability coefficient = 0.83, gain = 0.74).

[0189] The system uses the optimal parameter vector p * The lifting platform's movement is controlled, and the platform's response is as follows: Actual lifting height: 2.95m; Attitude change rate Δθ = 1.28° / s; Control delay time T delay =0.92s; compensation error ∈ comp =0.34°.

[0190] According to the labeling rules, the system assesses this round of control as "normal operation". Response data r 2218 ={H,Δθ,T delay ,∈ comp The ,label} was added to the response dataset R exec It is used for subsequent model optimization and control strategy adjustment.

[0191] This embodiment demonstrates step by step how to start from sensor sampling, through multi-scale perturbation characterization, dynamic prior update, control space construction, secretary bird optimal search, and finally control issuance and execution feedback, showcasing the feasibility and effectiveness of this invention in practical engineering applications.

[0192] This invention proposes a multi-scale variational contrastive learning architecture for disturbance perception. By introducing scale energy distribution and a self-attention cross-scale collaborative mechanism, the model can extract the temporal variation patterns of multimodal information on acceleration, angular velocity, and significant wave height from disturbance response windows at different frequencies, and obtain stable representations by fusing variational sampling mechanisms. Based on this, a dynamic disturbance prior distribution network constructed from time series data is embedded, replacing the traditional fixed standard normal distribution. By predictively modeling the evolution trend of disturbance distribution, the model's sensitivity and stability to the temporal characteristics of sea state disturbances are enhanced. This effectively solves the problems of insufficient ability to identify unstructured strong noise signals and overly static disturbance prior assumptions in traditional disturbance feature extraction methods, significantly improving the accuracy of disturbance perception and system robustness.

[0193] This invention constructs a secretary bird population initialization strategy guided by the potential feature vector of disturbance. Combined with the dynamic generation of the initial control parameter location distribution by the disturbance intensity, the secretary bird population starts from the highly correlated region of the disturbance state, skipping the inefficient exploration process caused by a large number of random initializations in traditional algorithms. At the same time, by driving the step size adjustment coefficient through the disturbance intensity index, a two-way closed-loop mechanism of disturbance perception-optimization search is constructed. When the sea state changes drastically, it dynamically converges to a high-precision local region, and maintains global search capability under stable disturbance conditions. The improved strategy is significantly better than the traditional secretary bird and particle swarm optimization algorithms with fixed step size. In high-dimensional parameter space and nonlinear control function scenarios, the speed of control parameter optimization is improved.

[0194] This invention establishes a direct feedback channel between the disturbance perception model and the platform lifting and lowering execution system. The attitude change rate, wave height response, and control delay indices generated in real time during the lifting and lowering process constitute the platform lifting and lowering response dataset, with operating status labels defined according to disturbance compensation error and attitude stability. This dataset not only evaluates the effectiveness of the current control strategy but also uses structured feedback information to influence the dynamic prior prediction network and the secretary bird optimizer, enabling online updates to the control strategy and disturbance prior model, achieving intelligent adaptive control driven by disturbances. This interconnected closed-loop system significantly outperforms the traditional unidirectional "perception-control-execution" architecture, exhibiting extremely high dynamic adaptability and engineering practicality.

[0195] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A self-supervised learning-based sea state adaptive ascent and descent control method for rocket platforms, characterized in that, Includes the following steps: S1. A data acquisition system is deployed synchronously on the rocket platform to continuously acquire and preprocess sea state disturbance datasets, generating preprocessed sea state disturbance datasets. S2. Construct a multi-scale self-supervised variational contrastive learning model, and perform self-supervised training with a preprocessed sea state disturbance dataset as input, outputting a set of multi-scale disturbance feature representations; S3. Based on the multi-scale perturbation feature representation set, construct a dynamic prior distribution model of sea state perturbation of rocket platform, and embed the dynamic prior distribution model into multi-scale self-supervised variational contrastive learning model to complete model update, and obtain dynamic prior variational perturbation encoder; S4. Use a dynamic prior variational disturbance encoder to infer the current preprocessed sea state disturbance dataset, output the real-time sea state disturbance feature vector, and construct a multi-dimensional search space for lifting control parameters based on the real-time sea state disturbance feature vector and the multi-scale disturbance feature representation set, and set the control performance evaluation function. S5. The real-time sea state disturbance feature vector is used to guide the initial position of the secretary bird population. The dynamic guided secretary bird optimization algorithm is started to perform a global search in the multi-dimensional search space of the lifting control parameters. The optimal lifting control parameter set is obtained by iterative calculation through the dynamic guided secretary bird optimization algorithm. S6. Input the optimal lifting and lowering control parameter set into the rocket platform lifting and lowering execution system and complete the platform lifting and lowering control, generating platform lifting and lowering response data.

2. The method for adaptive sea state ascent and descent control of a rocket platform based on self-supervised learning according to claim 1, characterized in that, S1 includes the following steps: S11. By deploying accelerometers, gyroscopes, and sea state monitors at key structural locations on the rocket platform, and setting sampling time intervals and total sampling durations, disturbance signals of the rocket platform under dynamic sea state environments are continuously collected to form an original sea state disturbance dataset. Each disturbance sampling data point in the original sea state disturbance dataset... Including timestamps Three-dimensional acceleration vector 3D angular velocity vector and instantaneous effective wave height value The total number of disturbance data is ; S12. Perform data preprocessing on the original sea state disturbance dataset, using a time window length of [missing information]. The sliding filter method smooths the three-dimensional acceleration vector, three-dimensional angular velocity vector and instantaneous effective wave height value, eliminates high-frequency disturbance components caused by sensor sampling jitter, and performs anomaly detection and elimination of the signal based on the adaptive threshold mechanism, excludes atypical disturbance samples affected by external abnormal interference, and obtains the intermediate processing dataset after filtering and denoising. S13. Perform perturbation feature enhancement processing on the intermediate processing dataset, calculate the perturbation rate of change feature between continuous perturbation data based on the time step, and extract the three-dimensional acceleration rate of change. 3D angular velocity change rate With the rate of change of effective wave height These represent the acceleration fluctuation trend, attitude disturbance fluctuation trend, and sea state disturbance fluctuation trend of the rocket platform at the current moment relative to the previous moment, respectively. The disturbance rate of change characteristics are concatenated with the corresponding three-dimensional acceleration vector, three-dimensional angular velocity vector, and instantaneous significant wave height value to form an enhanced disturbance data sample set. This enhanced disturbance data sample set is used as the final preprocessed sea state disturbance dataset. .

3. The method for adaptive sea state ascent and descent control of a rocket platform based on self-supervised learning according to claim 2, characterized in that, S2 includes the following steps: S21. Construct a multi-scale self-supervised variational contrastive learning model to preprocess the sea state disturbance dataset. As input, for the multi-frequency oscillation characteristics contained in the perturbation sample, a scale set S is defined, where each scale in the scale set S... The response window corresponding to different frequency disturbance characteristics in the rocket platform's ascent and descent system is obtained through a scale filter. Applying this to the channel splicing feature vectors to construct scale perturbation feature signals ; S22. Calculate the energy of the scale perturbation characteristic signal to measure its potential impact on the lift-down stability of the rocket platform. Define the scale energy as: ； in, Indicates perturbation sample In scale The total energy at this scale represents the driving force of the disturbance signal on the dynamic response capability of the rocket platform. S23. Normalize the scale energy to obtain the scale energy weights of the perturbed samples. ; S24. For the perturbation characteristic signal at each scale Configure a parallel input global perturbation encoder With local disturbance encoder Slow-change trend disturbances and instantaneous disturbances were extracted during the ascent and descent of the rocket platform, respectively: ； ； in, Representing scale The mean vector of the lower perturbation characteristic distribution. For scale The standard deviation vector of the lower perturbation characteristic distribution; S25. Generate standard normal noise independently for each scale based on a reparameterization strategy. and By combining the corresponding mean vector and standard deviation vector, the long-term disturbance latent vector is obtained. With the latent vector of instantaneous perturbation ; S26. Calculate the long-term perturbation latent vector. With the latent vector of instantaneous perturbation The cosine similarity between them, combined with the dimension of latent variables. Temperature scaling is performed, and then the softmax function is applied to obtain the cross-scale attention coefficients. The cross-scale attention coefficient reflects the mutual driving relationship of perturbations at different scales and is used to construct a coordinated rise and fall control signal path. S27. According to scale energy weight and cross-scale attention coefficient For instantaneous disturbance latent vector Perform weighted summation to generate a perturbation fusion vector. The disturbance fusion vector integrates the combined effects of sea state disturbances at various scales on the rocket platform's ascent and descent at the current moment; S28. Fusion perturbation vector Input disturbance stability prediction head Output the predicted slope value of future wave height change. And based on the actual wave height variation trend of the rocket platform under the current disturbance. Compared with the predicted value The mean square error between them defines the stability prediction loss. It is used to measure the accuracy of predictions regarding the future stability of rocket platforms; S29. By jointly optimizing the multi-scale variational reconstruction loss Self-monitored comparison loss With stability prediction loss and with adjustment coefficient and By controlling the relative weights of various losses in the sea state disturbance perception task, the parameters of the multi-scale self-supervised variational contrastive learning model are updated: ； in, and These are the adjustment coefficients for contrastive learning and prediction losses, respectively, which control the relative weights of feature learning and stability judgment in the perturbation perception task; S210. After completing the joint training, output a set of multi-scale perturbation feature representations. .

4. The self-supervised learning-based sea state adaptive ascent and descent control method for rocket platforms according to claim 3, characterized in that, S3 includes the following steps: S31. Based on multi-scale perturbation feature representation set According to the timestamp of the perturbation sample Construct time-series trajectories of perturbation features and fuse each perturbation vector. Its corresponding timestamp Composition of perturbation feature time-series state trajectory set It is used to characterize the evolution characteristics of sea state disturbances of rocket platforms over a continuous time period; S32. In the perturbation characteristic time-series state trajectory set Set a fixed length sliding time window. The time series trajectory set of disturbance features is divided into several disturbance local sub-window sequences by using a sliding time window. Each disturbance local sub-window sequence contains a time period. The set of all fused perturbation feature vectors is denoted as the perturbation feature sub-window set. It is used to extract the local disturbance distribution pattern of the rocket platform at different stages of sea state change; S33. For each set of perturbation feature sub-windows The fused perturbation feature vectors are statistically analyzed, and the mean vector of all perturbation features in the perturbation feature sub-window set is calculated. With covariance matrix These are used to describe the trend of the disturbance center and the spread range of the disturbance characteristics of the rocket platform in the current time period, respectively. S34. Constructing a Prior Prediction Network timestamp of the perturbation sample and corresponding perturbation fusion vector Input it, output the estimated prior distribution parameters To estimate the potential response structure of the rocket platform to sea state disturbance at the current moment, and to define the prior estimation error loss function using KL divergence: ； in, These are the posterior distribution parameters calculated by the multi-scale encoder at the current time. The KL divergence measures the structural deviation between the current state of the platform and the predicted perturbation trend, which is the estimated prior distribution parameter of the prior prediction network output. S35. Estimate the prior distribution parameters from the output of the prior prediction network. Embedded in a multi-scale self-supervised variational contrastive learning model, it replaces the standard normal distribution as a new variational prior distribution, completing the dynamic structural update of the disturbance perception model during the rocket platform's ascent and descent, thus forming a dynamic prior variational disturbance encoder.

5. The self-supervised learning-based sea state adaptive ascent and descent control method for rocket platforms according to claim 4, characterized in that, S4 includes the following steps: S41. Based on the dynamic prior variational disturbance encoder, receive the latest enhanced disturbance data sample from the current preprocessed sea state disturbance dataset. And with its channel splicing characteristics Using this as input, infer the posterior distribution parameters of the perturbation at the current time step. Simultaneously, the prior prediction network generates the current prior distribution parameters. Based on a dynamic prior distribution, a reparameterization strategy is used to generate the real-time perturbation latent feature vector at the current moment: ； in, This represents the real-time potential perturbation feature vector of the platform in the multi-scale perturbation space at the current moment. This is a standard normal noise vector, representing an abstract representation of the current sea state disturbance during the rocket platform's ascent and descent. S42. Perturb the latent feature vector in real time. With multi-scale perturbation feature representation set Scale alignment is performed by using temperature coefficients. Attention weights are obtained using a modulated cosine similarity-softmax mechanism. Then, the attention weights are applied to the perturbation fusion vector. Weighted summation yields the perturbation-driven fusion vector. ; S43. Based on the perturbation-driven fusion vector Constructing a multi-dimensional search space for lifting control parameters The multidimensional search space for lifting control parameters is composed of the control parameter vector. Composition, each control parameter in the control parameter vector All are constrained to the corresponding minimum value With the maximum value between; S44. Multidimensional search space for lifting control parameters Define control performance evaluation function The control performance evaluation function is composed of highly fluctuating residuals. Attitude stability error and control response delay time The sum of three weighted squares is used, with the weighting coefficients as follows: , , It is used to comprehensively measure the lift control accuracy and response efficiency of the rocket platform under the current potential disturbance state.

6. The method for adaptive sea state ascent and descent control of a rocket platform based on self-supervised learning according to claim 5, characterized in that, S5 includes the following steps: S51. Based on the real-time perturbation latent feature vector Calculate the current disturbance intensity index Among them, the disturbance intensity index Based on real-time perturbation latent feature vectors The L2 norm is determined and used to measure the overall energy amplitude of the disturbance experienced by the rocket platform under current sea conditions; S52. In the multi-dimensional search space of lifting control parameters Built-in includes Individual secretary bird populations, and real-time perturbation of latent feature vectors. For heuristic information, through the weight matrix With bias vector After linear transformation and compression using the Sigmoid function, the perturbation mapping coefficients are obtained. Then according to the lower bound vector With upper limit vector The interval proportional mapping method generates the initial control parameter vector for each secretary bird. Complete the perturbation-driven initialization of the secretary bird population; S53. Based on the disturbance intensity index Set adaptive step size adjustment factor Adaptive step size adjustment factor Based on the basic step size With step size decay coefficient A joint decision is made when the disturbance intensity index When increased, the adaptive step size adjustment factor Decrease according to the inverse proportional relationship; S54. During the Secretary Bird optimization iteration process, for the first... Each individual secretary bird in the generation updates its current control parameter vector using a jump-tracking update strategy. Add adaptive step size adjustment factor With random weights , Hybrid directional increments achieve the goal of moving towards the optimal control parameter vector. With random individual vectors Simultaneously converge, thereby obtaining the next generation control parameter vector. : ； in, Indicates the first In the middle control performance evaluation function The optimal control parameter vector, Let be a vector of random individuals in the current population. In order to be in Random weights for independent uniform sampling within intervals are used to simulate the hunting focus behavior and global reconnaissance behavior of the secretary bird. S55. Repeat the secretary bird population update until the set termination condition is met. The termination condition includes a maximum iteration limit or a control performance evaluation function for a certain number of consecutive generations. If the improvement is less than the threshold, the final output will be the control performance evaluation function that performs best throughout the entire iteration process. Control parameter vector that achieves minimum value Control parameter vector This serves as the current optimal set of lifting control parameters for the rocket platform's lifting and lowering execution system.

7. The method for adaptive sea state ascent and descent control of a rocket platform based on self-supervised learning according to claim 1, characterized in that, S6 includes the following steps: S61. Transfer the optimal lift control parameter vector Input the rocket platform's ascent and descent execution system to drive the platform to execute the current ascent and descent task. The ascent and descent execution system adjusts the ascent and descent rate, attitude angle stability factor, and disturbance compensation gain key control signals according to the strategy values ​​corresponding to each control dimension in the optimal control parameter vector. S62. During the ascent and descent process, a sensor system fixedly connected to the rocket platform structure monitors the dynamic process in real time, recording the platform's altitude change curve, attitude change trajectory, attitude deviation rate, and ascent and descent time delay platform response variables within the ascent and descent response cycle, generating a platform ascent and descent response dataset. ; S63. Platform Height-Rise Response Dataset Each response data Includes the following parameter field: the platform's current elevation height. Platform attitude angle change rate Control command response time Disturbance compensation error System control status label System control status label The description uses labels to indicate the current operating status of the platform.

8. The method for adaptive sea state ascent and descent control of a rocket platform based on self-supervised learning according to claim 7, characterized in that, The label identifier is defined as follows: Normal operating condition: Disturbance compensation error Response time Rate of change of attitude angle ; Disturbed operating status: Disturbance compensation error Response time or rate of change of attitude angle ; Control instability: Disturbance compensation error Response time or rate of change of attitude angle .

Images