Sea rocket erecting state anomaly detection method based on sparse representation and adaptive filtering

Abstract

The application discloses a sea rocket erecting state anomaly detection method based on sparse representation and adaptive filtering, and comprises the following steps: acquiring state data in the rocket erecting process through a plurality of sensors, constructing a time sequence, setting a sliding window with change ability in each time period, and extracting sparse features for state prediction. By calculating the difference between the predicted value and the observed value, a feedback adjustment signal is formed to adjust the parameters in the state estimation and external disturbance modeling process respectively. A double-path structure is used to independently estimate the state and disturbance, and the final estimation result is obtained through a fusion strategy. The system continuously updates the parameters according to the error condition, realizes continuous adaptive adjustment and anomaly identification, improves the identification accuracy and response ability of the abnormal trend. The application is suitable for identifying and processing state anomalies in the rocket erecting process in the sea environment, and has strong dynamic adaptability and time sequence anomaly detection capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of spaceflight TT&C and intelligent signal processing technology, and particularly relates to a sea-based rocket erecting state anomaly detection method based on sparse representation and adaptive filtering. BACKGROUND

[0002] With the gradual normalization of sea-based space launch missions, the erecting process of the rocket in the complex sea state environment becomes an important link affecting the safety and success of the mission. Timely detection and accurate identification of the erecting state anomaly of the rocket are of great significance to ensure the structural stability, avoid abnormal deduction and improve the controllability of the erecting process. However, in the sea environment, there are uncertainties such as wind and wave disturbance, large sensor measurement noise and highly nonlinear structural response, which leads to significant challenges in the practical application of traditional state detection methods.

[0003] In the prior art, common state anomaly detection methods mostly rely on fixed threshold judgment, single-channel data monitoring or linear filtering estimation. These methods are difficult to effectively capture weak anomalies or dynamic trends when facing multi-source information fusion, disturbance dynamic changes and high-noise environments in the erecting process, and have the following defects:

[0004] 1. Static judgment mechanism lags behind: Traditional methods mostly use static thresholds or fixed discrimination rules, which cannot cope with nonlinear shifts and small trend variations in the state evolution process, and lack dynamic response capability.

[0005] 2. Weak multi-sensor information fusion capability: Existing methods are difficult to achieve collaborative modeling and compensation of multi-channel state information, especially when the data is incomplete or some sensors fail, the state estimation accuracy decreases significantly.

[0006] 3. Filter structure lacks adaptability: Traditional filtering methods do not distinguish between disturbance and state common mode signals, cannot distinguish between state evolution and disturbance response, and are prone to false estimation or delayed response.

[0007] 4. Single parameter adjustment mechanism: In the filtering and anomaly detection process, the parameters are mostly set manually, and there is a lack of adaptive adjustment strategies based on real-time feedback, making it difficult to adapt to non-stable systems under sea state disturbance.

[0008] 5. Limited anomaly recognition accuracy: In a high-interference, low signal-to-noise ratio background, traditional methods are difficult to accurately extract weak anomaly features from the residual, resulting in a high rate of missed or false judgments.

[0009] Therefore, how to provide a sea-based rocket erecting state anomaly detection method based on sparse representation and adaptive filtering is a problem that those skilled in the art need to solve. SUMMARY

[0010] One purpose of the present application is to propose a sea rocket erecting state anomaly detection method based on sparse representation and adaptive filtering, the present application constructs a multi-channel dynamic sliding window mechanism, combines sparse coding to extract state features, and introduces a state prediction and disturbance compensation double-path filtering structure, realizes adaptive adjustment of filtering gain and modeling parameters through residual feedback adjustment, and has the advantages of high identification accuracy, strong dynamic adaptation and real-time anomaly detection.

[0011] The sea rocket erecting state anomaly detection method based on sparse representation and adaptive filtering according to the embodiment of the present application comprises the following steps:

[0012] S1, acquiring original state observation data collected by a plurality of state monitoring sensors during a sea rocket erecting process, performing time alignment processing, and generating a multi-channel state observation time sequence;

[0013] S2, constructing a dynamic sliding window mechanism according to the multi-channel state observation time sequence, setting an adaptive window length for each dynamic sliding window, and generating a dynamic sparse representation dictionary;

[0014] S3, sparse coding of state observation data in the dynamic sliding window based on the dynamic sparse representation dictionary, extracting a state sparse feature vector, and calculating a current state prediction value according to the state sparse feature vector;

[0015] S4, calculating a residual signal between the current state prediction value and the actual state observation value, and calculating a feedback adjustment coefficient according to the residual signal;

[0016] S5, constructing a double-path filtering architecture containing a state prediction path and a disturbance compensation path, the state prediction path receiving the state sparse feature vector for performing state estimation, the disturbance compensation path receiving the residual signal for performing disturbance estimation, and fusing the outputs of the double paths to generate a final state estimation result;

[0017] S6, according to the feedback adjustment coefficient, linkage adjustment of the filtering gain parameter used when performing state estimation and the disturbance modeling parameter used when performing disturbance estimation, and construction of a parameter adaptive adjustment mechanism;

[0018] S7, inputting the final state estimation result and the residual signal into an anomaly detection model, the anomaly detection model identifying state anomalies in the sea rocket erecting process based on a time series consistency analysis method and outputting an anomaly labeling result.

[0019] Optionally, the S2 specifically comprises:

[0020] S21, performing unified time axis mapping on the multi-channel state observation time sequence, and constructing a multi-channel observation matrix covering synchronous data of each channel;

[0021] S22, setting an initial position of a sliding window in the multi-channel observation matrix, and extracting state observation data of each channel in a corresponding time period as a data input segment corresponding to the current sliding window;

[0022] S23, calculating a state change rate of each channel in the time period for the data input segment corresponding to the current sliding window, and constructing a nonlinear window length adjustment function according to an average value of the state change rates of all channels, and defining a current window length L(t) as:

[0023]

[0024] wherein, L min is a minimum length of the sliding window, L max is a maximum length of the sliding window, v i (t) is a state change rate of the i th channel in the current time period, and N is a total number of state monitoring channels;

[0025] S24, calculating a sliding residual signal between a state prediction value and an actual state observation value in a sliding cycle of the current sliding window, calculating a change amplitude Δε of the sliding residual signal, if the change amplitude Δε is greater than a set residual change threshold θ, the starting position of the sliding window is moved by a sliding step Δt, the position of the sliding window is updated, if the Δε is less than or equal to the residual change threshold θ, the starting position of the sliding window is kept unchanged;

[0026] S25, extracting an observation data segment in each sliding window triggered, and constructing a set of dynamic sparse representation dictionaries corresponding to the time characteristics of each sliding window.

[0027] Optionally, the S3 specifically comprises:

[0028] S31, extracting a state observation data segment in a time period corresponding to each sliding window, and constructing a to-be-encoded data matrix, each column of the to-be-encoded data matrix corresponding to a time sequence sample of a state monitoring channel;

[0029] S32, taking the to-be-encoded data matrix as an input, combining the dynamic sparse representation dictionary corresponding to the sliding window, and using an orthogonal matching pursuit algorithm to perform sparse coding on each state observation data, solving a sparse coefficient vector α, satisfying the following expression: x≈D·α, wherein x is a state observation vector, D is a dynamic sparse representation dictionary, α is a sparse coding coefficient vector, and a state sparse feature vector is extracted;

[0030] S33, inputting the sparse coding coefficient vector α into a state prediction structure, the state prediction structure being composed of two nonlinear mapping functions, including a gating activation function and a state projection function:

[0031]

[0032] wherein, is the current state prediction value, W g , W p are weight matrices of the gating path and the prediction path respectively, b g , b p are corresponding bias terms, and σ(·) is a nonlinear activation function.

[0033] S34, arranging the state prediction values generated by the continuous sliding window in chronological order to construct a state prediction sequence.

[0034] Optionally, the S4 specifically includes:

[0035] S41, obtaining a current state prediction value in the constructed state prediction sequence and an actual state observation value y t at a corresponding time as a current prediction value and a current observation value respectively.

[0036] S42, calculating a residual signal ε t according to a difference between the current state prediction value and the actual state observation value y t .

[0037] S43, calculating a residual change rate Δε t based on a residual signal sequence {ε t-n ,…,ε t} of a plurality of continuous time points, the residual change rate being a difference between a current time residual signal and a last time residual signal.

[0038] S44, constructing a feedback adjustment function f(·) according to the residual signal ε t and the residual change rate Δε t , and calculating a feedback adjustment coefficient γ t , the feedback adjustment function adopting an exponential weight combination form and being expressed as follows:

[0039] γ t = exp(-λ1|ε t |-λ2|Δε t |);

[0040] wherein, γ t is the feedback adjustment coefficient, λ1 and λ2 are respectively adjustment weight coefficients of a residual amplitude and a change rate, and are used for controlling a response degree to a residual intensity and a change trend.

[0041] Optionally, the S5 specifically includes:

[0042] S51, receive the state sparse feature vector, take the state sparse feature vector as the input of the state prediction path, and use the prediction model in the Kalman filter structure to perform state estimation to generate a state estimation value:

[0043]

[0044] wherein, is the current state estimation value, is the previous state estimation value, A is a state transition matrix, B is a control input matrix, u t is an optional control input;

[0045] S52, receive the residual signal, take the residual signal as the input of the disturbance compensation path, and use the exponential weighted residual modeling method to perform disturbance estimation to generate a disturbance estimation value:

[0046]

[0047] wherein, d t is the current disturbance estimation value, ε t-i is the residual signal before the i-th step, ω i is a disturbance estimation weight coefficient, satisfying a normalization condition;

[0048] S53, weight and fuse the state estimation value generated by the state prediction path and the disturbance estimation value generated by the disturbance compensation path to construct a final state estimation value, and the fusion method uses a weighted average model, which is represented as:

[0049]

[0050] wherein, is the current final state estimation value, and λ is a fusion weight coefficient, which balances and adjusts between the state estimation and the disturbance estimation.

[0051] Optionally, the S6 specifically includes:

[0052] S61, receive the feedback adjustment coefficient, and input the feedback adjustment coefficient to the state prediction path and the disturbance compensation path respectively, to link and adjust the filter gain parameter in the state estimation path and the disturbance modeling parameter in the disturbance compensation path;

[0053] S62, in the state prediction path, dynamically update the filter gain parameter according to the value of the current feedback adjustment coefficient, the filter gain parameter is used for the state update step in the Kalman filter structure, when the feedback adjustment coefficient increases, the filter gain parameter is correspondingly increased, to enhance the tracking ability of the rapid change of the state, when the feedback adjustment coefficient decreases, the filter gain parameter is correspondingly reduced, to suppress the invalid response;

[0054] S63, in the disturbance compensation path, adjusting the disturbance modeling parameters according to the change trend of the current feedback adjustment coefficient, the disturbance modeling parameters including the time weighting window length of the disturbance estimation and the weighted distribution structure of the historical residual signal, the proportion of the current residual signal in the disturbance estimation is strengthened when the feedback adjustment coefficient increases, and the influence degree of the historical disturbance trend information is improved when the feedback adjustment coefficient decreases;

[0055] S64, in the current sliding window period, respectively calculating the state estimation error between the state estimation value of the state prediction path output and the actual state observation value, and the disturbance estimation error between the disturbance estimation value of the disturbance compensation path output and the actual state observation value, constructing a joint error evaluation index, and establishing a parameter adjustment criterion with the minimum joint error as the target;

[0056] S65, according to the output result of the joint error evaluation index, dynamically adjusting the adjustment step and update direction of the filter gain parameter used in the state prediction path and the disturbance modeling parameter used in the disturbance compensation path, and constructing a double-factor parameter adaptive adjustment mechanism coupled with linkage;

[0057] S66, applying the filter gain parameter and the disturbance modeling parameter updated by the joint error to the execution process of the state prediction path and the disturbance compensation path in the next moment, forming a continuous parameter self-optimization closed loop system driven by the feedback adjustment coefficient as the core.

[0058] Optionally, S65 specifically includes:

[0059] S651, based on the calculated state estimation error and disturbance estimation error, generating a joint error evaluation index, the joint error evaluation index being composed of the weighted combination of the state estimation error and the disturbance estimation error;

[0060] S652, comparing the joint error evaluation index with a preset joint error threshold, and judging whether to start the parameter adjustment process according to the comparison result, if the joint error evaluation index is greater than the preset joint error threshold, the parameter adjustment is performed, and if the condition is not met, the current parameter setting is maintained;

[0061] S653, in the parameter adjustment, the change value of the state estimation error is called as the adjustment control input to the state prediction path, and the filter gain parameter is adjusted, the adjustment operation sets the adjustment step and direction according to the error change value, and updates the filter gain parameter;

[0062] S654, the change value of the disturbance estimation error is called synchronously as the control input transmitted to the disturbance compensation path, for adjusting the disturbance modeling parameter, the disturbance modeling parameter adjustment strategy including adjusting the time weighting distribution and estimation window span of the residual signal;

[0063] S655, jointly analyze the error change value of the state prediction path and the error change value of the disturbance compensation path, calculate a joint adjustment proportion factor, and the joint adjustment proportion factor is used to configure an adjustment resource between the two paths;

[0064] S656, output according to the updated filter gain parameter and the disturbance modeling parameter respectively, and pass to S66 to build a continuous double-path parameter linkage update chain.

[0065] The beneficial effects of the present application are:

[0066] (1) The present application realizes fine-grained extraction and dynamic modeling of the sea rocket erecting state characteristics by introducing a dynamic sliding window mechanism and a sparse representation method, combining a nonlinear gating prediction structure and an orthogonal matching pursuit algorithm, and significantly improves the state recognition ability in a high noise and strong disturbance background. The linkage update process of the filter gain parameter and the disturbance modeling parameter is constructed based on the feedback regulation mechanism of the residual error, and a double-factor coupled regulation mechanism is established between the state prediction path and the disturbance compensation path, so that the system has real-time parameter optimization and structure adaptive ability, thereby improving the stability of state estimation and the accuracy of abnormal detection.

[0067] (2) The present application realizes dynamic collaborative control of the filter path and the disturbance path by using the parameter adjustment criterion driven by the joint error evaluation index, and ensures the continuous tracking ability of the algorithm to the small abnormal trend by constructing the residual error activation and sliding window adjustment strategy, effectively reducing the misjudgment rate caused by sensor drift or incomplete data. Combined with the time sequence consistency determination method of the abnormal detection model, the present method can realize multi-dimensional abnormal perception in the whole process of rocket erecting, and label and identify the potential fault trend in advance.

[0068] (3) The present application realizes the structured data receiving, feature extraction, prediction estimation, disturbance compensation, adjustment optimization and abnormality determination closed-loop control in the whole erecting process state recognition and abnormality monitoring process, does not depend on static threshold or single model response, has the cross-model self-update, self-regulation and weak signal recognition ability, and is suitable for high-robustness state perception tasks in complex sea conditions. BRIEF DESCRIPTION OF DRAWINGS

[0069] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0070] Figure 1 The flowchart of the sea rocket erecting state abnormality detection method based on sparse representation and adaptive filtering proposed by the present application. DETAILED DESCRIPTION

[0071] The application will be described in further detail below with reference to the drawings. These drawings are simplified schematic diagrams and only show the basic structure of the application in a schematic manner, and thus only show the components relevant to the application.

[0072] Reference Figure 1 The offshore rocket erecting state anomaly detection method based on sparse representation and adaptive filtering comprises the following steps:

[0073] S1, acquiring original state observation data collected by multiple state monitoring sensors during the erecting process of the offshore rocket, performing time alignment processing, and generating a multi-channel state observation time sequence;

[0074] In this embodiment, multiple state monitoring sensors are arranged on the offshore rocket erecting platform to collect real-time multi-dimensional original state observation data of the rocket structure, attitude, environmental disturbance, etc. Due to the differences in sampling frequency and time stamp offset of each sensor, the original data is processed by time alignment processing using interpolation resampling and unified time reference alignment algorithm, an observation matrix on a unified time axis is constructed, and the comparability of the data in each channel at the same time point is ensured. The multi-channel state observation time sequence formed finally provides basic data support for subsequent sliding window division, sparse coding and anomaly detection, realizes effective conversion of multi-source asynchronous data into structured time sequence input, and improves the consistency and accuracy of state feature extraction.

[0075] S2, constructing a dynamic sliding window mechanism according to the multi-channel state observation time sequence, setting an adaptive window length for each dynamic sliding window, and generating a dynamic sparse representation dictionary;

[0076] S3, performing sparse coding on the state observation data in the dynamic sliding window based on the dynamic sparse representation dictionary, extracting a state sparse feature vector, and calculating a current state prediction value according to the state sparse feature vector;

[0077] S4, calculating a residual signal between the current state prediction value and the actual state observation value, and calculating a feedback adjustment coefficient according to the residual signal;

[0078] S5, constructing a double-path filtering architecture comprising a state prediction path and a disturbance compensation path, the state prediction path receiving the state sparse feature vector for performing state estimation, the disturbance compensation path receiving the residual signal for performing disturbance estimation, and fusing the outputs of the double paths to generate a final state estimation result;

[0079] S6, according to the feedback adjustment coefficient, linkage adjustment of the filtering gain parameter used when performing state estimation and the disturbance modeling parameter used when performing disturbance estimation, and construction of a parameter adaptive adjustment mechanism;

[0080] S7, input the final state estimation result and the residual signal into an anomaly detection model, and the anomaly detection model identifies state anomalies in the sea rocket erecting process based on a time series consistency analysis method and outputs an anomaly labeling result.

[0081] The embodiment takes the final state estimation result and the residual signal as input, constructs an anomaly detection model, extracts the continuity, trend deviation and residual mutation characteristics of the state estimation sequence and the residual signal sequence within a time window through joint analysis of the time evolution consistency between the state estimation sequence and the residual signal sequence, and identifies abnormal change patterns in the erecting process by using a dynamic threshold perception mechanism and a sliding correlation matching method. When significant inconsistency exists between the predicted state and the observed data, and the inconsistency continues to exist in multiple time steps, the model automatically determines that it is a potential anomaly and outputs the corresponding anomaly labeling result. The method realizes multi-dimensional perception of weak variation trend and sudden anomaly, and improves the accuracy and stability of the rocket erecting state anomaly recognition.

[0082] In the embodiment, the S2 specifically comprises:

[0083] S21, performing unified time axis mapping on the multi-channel state observation time sequence, and constructing a multi-channel observation matrix covering synchronous data of each channel;

[0084] S22, setting an initial position of a sliding window in the multi-channel observation matrix, and extracting state observation data of each channel in a corresponding time period as a data input segment corresponding to the current sliding window;

[0085] S23, for the data input segment corresponding to the current sliding window, calculating a state change rate of each channel in the time period, and constructing a nonlinear window length adjustment function according to an average value of the state change rates of all channels, and defining a current window length L(t) as:

[0086]

[0087] wherein, L min is the minimum length of the sliding window, L max is the maximum length of the sliding window, v i (t) is the state change rate of the i th channel in the current time period, and N is the total number of state monitoring channels;

[0088] The formula uses a hyperbolic tangent function to construct a continuous and derivable nonlinear response curve, so that when the state change rate is small, the window length remains at a basic level, and when the change rate significantly increases, the window length gradually shrinks within a limited range, thereby enhancing the sensitivity of the system to state mutations. The design effectively balances the stability and response speed of feature extraction, adaptively adjusts the sliding window granularity in different state fluctuation environments, and improves the timeliness and accuracy of sparse feature representation.

[0089] S24, calculate the sliding residual signal between the state prediction value and the actual state observation value in a sliding period on the current sliding window, calculate the change amplitude Δε of the sliding residual signal, if the change amplitude Δε is greater than the set residual change threshold θ, then the starting position of the sliding window is moved by one sliding step Δt, the position of the sliding window is updated, if Δε is less than or equal to the residual change threshold θ, the starting position of the sliding window is kept unchanged;

[0090] S25, extract the observation data segment in the window triggered to slide each time, and construct a set of dynamic sparse representation dictionaries corresponding to the time characteristics for each sliding window.

[0091] The embodiment maps the multi-channel state observation time sequence to a unified time axis, constructs a synchronous observation matrix, and sets a sliding window on the matrix. The state change rate of each data segment in the window is calculated. A nonlinear function is used to adaptively adjust the window length, and the sliding residual signal change amplitude is introduced as a dynamic trigger condition for window sliding, so that the sliding mechanism has responsiveness and flexibility. After the window sliding is triggered, the corresponding observation data is extracted and a dynamic sparse representation dictionary matching the time characteristics is constructed, so that the local state change is efficiently expressed and modeled. The embodiment dynamically adjusts the sliding window structure and the sparse dictionary update mechanism, improves the time sequence adaptability and feature expression accuracy of the method in the continuous modeling process of the vertical state, and lays a stable data processing foundation for subsequent state prediction and anomaly detection.

[0092] In the embodiment, the S3 specifically comprises:

[0093] S31, extract the state observation data segment in the time period corresponding to each sliding window, and construct a to-be-encoded data matrix. Each column of the to-be-encoded data matrix corresponds to a time sequence sample of a state monitoring channel.

[0094] S32, take the to-be-encoded data matrix as input, combine the dynamic sparse representation dictionary corresponding to the sliding window, and use the orthogonal matching pursuit algorithm to perform sparse coding on each state observation data to solve the sparse coefficient vector α, which satisfies the following expression: x≈D·α, wherein x is the state observation vector, D is the dynamic sparse representation dictionary, α is the sparse coding coefficient vector, and the state sparse feature vector is extracted.

[0095] S33, input the sparse coding coefficient vector α into the state prediction structure, which is composed of two nonlinear mapping functions, including a gating activation function and a state projection function:

[0096]

[0097] wherein, W is the predicted value for the current state. g W p These are the weight matrices for the gated path and the predicted path, respectively. g b p For the corresponding bias term, σ(·) is the nonlinear activation function;

[0098] The formula reflects that the state prediction is the product of the results from two independent paths. One path modulates the response of sparse features through a nonlinear activation function, while the other path obtains the basic prediction through linear projection. The gating path is responsible for adjusting the activation level of different features in the current state, while the projection path provides the original state mapping output. The fusion of the two can effectively capture the nonlinear relationships between features, achieve fine prediction of complex dynamic states, and improve the ability to model subtle state evolution trends.

[0099] S34. Arrange the state prediction values ​​generated by the continuous sliding window in chronological order to construct a state prediction sequence.

[0100] This implementation extracts multi-channel state observation data within a sliding window to construct a data matrix to be encoded. Combining this with a dynamic sparse representation dictionary, an orthogonal matching pursuit algorithm is used to sparsely encode the state observation vectors, yielding a sparse state feature vector. This feature vector is then input into a nonlinear state prediction structure composed of a gated activation function and a state projection function to generate the current state prediction value and construct a state prediction sequence. This approach introduces sparse reconstruction and nonlinear prediction mechanisms into the state modeling process, preserving key dynamic changes in the rocket's erection state while avoiding overfitting and miscalculation risks. This significantly improves the accuracy of state prediction under complex disturbance conditions and the ability to respond to minor anomalies.

[0101] In this embodiment, S4 specifically includes:

[0102] S41. Obtain the current state prediction value from the constructed state prediction sequence. and the actual state observation value y at the corresponding time. t , respectively, are used as the current predicted value and the current observed value;

[0103] S42. Predict the value based on the current state. Compared with the actual state observation value y t Calculate the residual signal ε based on the difference between them. t ;

[0104] S43, Based on the residual signal sequence {ε} at multiple consecutive time points t-n ,…,ε t} Calculate the residual rate of change Δε t The residual change rate is the difference between the residual signal at the current time and the residual signal at the previous time.

[0105] S44、according to the residual signal ε t with the residual change rate Δε t , a feedback adjustment function f(·) is constructed to calculate a feedback adjustment coefficient γ t , the feedback adjustment function adopts an exponential weight combination form, expressed as follows:

[0106] γ t = exp(-λ1|ε t |-λ2|Δε t |);

[0107] wherein γ t is the feedback adjustment coefficient, λ1 and λ2 are respectively the adjustment weight coefficients of the residual amplitude and the change rate, used to control the response degree of the residual intensity and the change trend.

[0108] The formula adopts an exponential decay structure, which dynamically generates an adjustment coefficient between zero and one according to the amplitude of the current residual signal and the size of the residual change rate. The coefficient is used to control the adjustment amplitude and direction of the subsequent parameters, to realize the synchronous response adjustment of the state prediction path and the disturbance compensation path. The exponential structure in the formula can effectively suppress the violent fluctuation of the adjustment coefficient caused by abnormal values, so that the adjustment behavior has smoothness and gradualness, has strong numerical stability and dynamic adaptive ability, so as to ensure that the parameter update has sufficient sensitivity in the early stage of abnormality, and gradually converges when the system tends to be stable, avoiding over-correction.

[0109] The embodiment constructs a residual signal by obtaining the difference between the current state prediction value and the actual state observation value, and further calculates a residual change rate by combining the change of the residual signal in the continuous time period, to form the state deviation characteristics from the time evolution perspective. On this basis, an exponential weight type feedback adjustment function is constructed by using the residual signal and the residual change rate as inputs to generate a feedback adjustment coefficient. The feedback adjustment coefficient, as the core variable of the system adaptive control, links and affects the adjustment of the subsequent filter gain parameters and disturbance modeling parameters. By introducing the dynamic combination mechanism of the residual information intensity and the trend change, the sensitive response and real-time feedback to the abnormal deviation evolution process are realized, and the identification accuracy and control flexibility of the system to the nonlinear deviation and disturbance mutation in the sea rocket erecting process are enhanced.

[0110] In the embodiment, the S5 specifically comprises:

[0111] S51, receiving a state sparse feature vector, taking the state sparse feature vector as the input of the state prediction path, and using the prediction model in the Kalman filter structure to generate a state estimation value for state estimation:

[0112]

[0113] wherein, is the current state estimate, is the previous state estimate, A is the state transition matrix, B is the control input matrix, u t is the optional control input;

[0114] The formula reflects the evolution relationship of the system state in the time dimension, wherein the state transition matrix is used to describe the state migration characteristics of the system under the condition of no disturbance, and the control input matrix reflects the influence of external input on the state change. The prediction process provides an initial estimate for the subsequent state update and error correction, ensures the time consistency and recursion of the entire filtering process, and is the basic link to realize state tracking and continuous estimation.

[0115] S52, receive the residual signal, take the residual signal as the input of the disturbance compensation path, and use the exponentially weighted residual modeling method to estimate the disturbance to generate a disturbance estimate:

[0116]

[0117] wherein, d t is the current disturbance estimate, ε t-i is the residual signal before the i-th step, ω i is the disturbance estimation weight coefficient, satisfying the normalization condition;

[0118] This method gives different weights to the residual signals of multiple time steps in history to form a weighted sum of the disturbance estimate, and the weight coefficient satisfies the normalization constraint. The principle of the formula is to use the principle of time proximity to make the residual signals of the closer time steps have a greater impact on the current disturbance estimate, thereby enhancing the response ability of the model to sudden disturbances, and improving the accuracy of the description of the disturbance trend while maintaining the stability of the estimate.

[0119] S53, weight and fuse the state estimate value generated by the state prediction path and the disturbance estimate value generated by the disturbance compensation path to construct the final state estimate value, and the fusion method uses a weighted average model, which is expressed as:

[0120]

[0121] wherein, is the current final state estimate, and λ is the fusion weight coefficient, which balances and adjusts between the state estimate and the disturbance estimate.

[0122] The formula realizes numerical fusion processing between the predicted state value of the state prediction path output and the disturbance estimation value of the disturbance compensation path output. The fusion mode generates the final state estimation value in a linear combination manner by setting a fusion weight coefficient and distributing the weight proportion between the two path outputs. The principle of the formula is that when the system state is less affected by the disturbance, the output of the state prediction path accounts for a relatively higher proportion; when the disturbance signal is more significant, the estimation result of the disturbance compensation path dominates in the fusion, thereby realizing a balanced mechanism of simultaneously modeling and responding to state changes and disturbance characteristics. The method realizes the integration of multi-path information without complex operations by using a simple and effective linear fusion strategy, thereby improving the stability and expression integrity of the state estimation result.

[0123] In the embodiment, the state sparse feature vector is input to the state prediction path, the current state is predicted based on a Kalman filter structure, the residual signal is input to the disturbance compensation path, the influence of the disturbance is estimated through an exponential weighting model, the state estimation value and the disturbance estimation value are generated by the two paths respectively, and then weighted fusion is performed to output the final state estimation result. The method realizes decoupled modeling of state evolution and external disturbance, effectively balances the influence of the two on the estimation result through the fusion strategy, improves the state tracking accuracy and system stability in complex sea conditions, and provides a reliable foundation for subsequent anomaly detection.

[0124] In the embodiment, the S6 specifically includes:

[0125] S61, a feedback adjustment coefficient is received, and the feedback adjustment coefficient is input to the state prediction path and the disturbance compensation path respectively, for linkage adjustment of a filter gain parameter in the state estimation path and a disturbance modeling parameter in the disturbance compensation path;

[0126] S62, in the state prediction path, the filter gain parameter is dynamically updated according to the value of the current feedback adjustment coefficient, the filter gain parameter is used for a state update step in the Kalman filter structure, the filter gain parameter is increased when the feedback adjustment coefficient is increased, so as to enhance the tracking ability of the rapid change of the state, and the filter gain parameter is reduced when the feedback adjustment coefficient is reduced, so as to suppress invalid response;

[0127] S63, in the disturbance compensation path, the disturbance modeling parameter is adjusted according to the change trend of the current feedback adjustment coefficient, the disturbance modeling parameter includes a time weighting window length of disturbance estimation and a weighted distribution structure of historical residual signals, the proportion of the current residual signal in disturbance estimation is increased when the feedback adjustment coefficient is increased, and the influence degree of historical disturbance trend information is increased when the feedback adjustment coefficient is reduced;

[0128] S64, calculate state estimation errors between state estimation values of state prediction path output and actual state observation values, and disturbance estimation errors between disturbance estimation values of disturbance compensation path output and actual state observation values respectively in the current sliding window period, construct a joint error evaluation index, and establish a parameter adjustment criterion with the goal of minimizing the joint error;

[0129] S65, dynamically adjust the adjustment step and update direction of the filter gain parameter used in the state prediction path and the disturbance modeling parameter used in the disturbance compensation path according to the output result of the joint error evaluation index, and construct a dual-factor parameter self-adaptive adjustment mechanism with linkage and coupling;

[0130] S66, apply the filter gain parameter and disturbance modeling parameter updated by the joint error to the execution process of the state prediction path and the disturbance compensation path in the next time, form a continuous parameter self-optimization closed loop system driven by the feedback adjustment coefficient as the core.

[0131] In this embodiment, the filter gain parameter in the state prediction path and the disturbance modeling parameter in the disturbance compensation path are adjusted based on the feedback adjustment coefficient. By introducing the joint error evaluation index, a dynamic weighting mechanism of state estimation error and disturbance estimation error is established, and linkage adjustment and coupling control of dual-path parameters are realized. The parameter adjustment step and update direction are corrected in real time in the sliding window period, and the updated parameters are continuously applied to the state estimation and disturbance compensation process in the next period, thereby constructing a feedback-driven parameter self-optimization closed loop system. This mechanism can dynamically adapt to system changes under the conditions of multiple source disturbances and non-stationary states, improve the stability of state estimation and the response ability to abnormal trends, and effectively enhance the adaptability and robustness of the state perception system during the rocket erecting process.

[0132] In this embodiment, S65 specifically includes:

[0133] S651, based on the calculated state estimation error and disturbance estimation error, generate a joint error evaluation index, which is composed of the weighted combination of the state estimation error and the disturbance estimation error;

[0134] S652, compare the joint error evaluation index with the preset joint error threshold, and determine whether to start the parameter adjustment process according to the comparison result. If the joint error evaluation index is greater than the preset joint error threshold, the parameter adjustment is performed, and if the condition is not met, the current parameter setting is maintained;

[0135] S653, in the parameter adjustment, the change value of the state estimation error is called as the adjustment control amount input to the state prediction path, and the filter gain parameter is adjusted. The adjustment operation sets the adjustment step and direction according to the error change value, and updates the filter gain parameter;

[0136] S654, synchronously call the change value of the disturbance estimation error as a control input to pass to the disturbance compensation path for adjusting the disturbance modeling parameters, and the disturbance modeling parameter adjustment strategy includes adjusting the time weighting distribution of the residual signal and the estimation window span;

[0137] S655, jointly analyze the error change value of the state prediction path and the error change value of the disturbance compensation path, calculate a joint adjustment proportion factor, and the joint adjustment proportion factor is used to configure the adjustment resources between the two paths;

[0138] S656, output the updated filter gain parameters and disturbance modeling parameters respectively, and pass to S66 to build a continuous double-path parameter linkage update chain.

[0139] The embodiment extracts the state estimation error and the disturbance estimation error, builds a joint error evaluation index, and judges whether to enter the parameter adjustment stage in combination with the preset threshold value, and realizes the adaptive control based on error dynamic feedback. In the adjustment process, the change value of the state estimation error and the disturbance estimation error is called respectively, the update step and direction of the filter gain parameters and the disturbance modeling parameters are controlled, and a joint adjustment proportion factor is introduced to configure the parameter adjustment resources between the state prediction path and the disturbance compensation path. The method forms a continuous double-path parameter linkage update chain, not only realizes the cooperative adjustment of state estimation and disturbance compensation, but also improves the adaptive ability of the system to parameter changes in a complex disturbance environment, and enhances the response accuracy and stability to abnormal states.

[0140] Example 1:

[0141] In order to verify the feasibility of the application in implementation, the application is applied to three sea rocket vertical lift tasks carried out by a certain coastal commercial launch platform in China from August 2024 to February 2025, and the test environment is located in the Yellow Sea. Due to the long-term interference of wind and wave and the influence of structural micro-vibration response of the launch platform, there is a significant uncertainty and time delay problem in the state monitoring during the rocket lifting process, and the conventional abnormal detection method based on static threshold and fixed filtering strategy cannot meet the real-time and accuracy requirements. Therefore, the system of the application is deployed in the platform state perception and control module, and real-time processing is carried out in combination with the data of 12 groups of multi-source inertial navigation, inclination, strain and acceleration sensors installed on the platform.

[0142] In the actual deployment process, the system first performs time series segmentation on each group of sensor data through a dynamic sliding window mechanism, automatically adjusts the window length and generates a multi-channel sparse feature dictionary, and uses the orthogonal matching pursuit algorithm to complete feature extraction. Then the current rocket attitude is estimated through a gated nonlinear prediction structure, and a feedback adjustment coefficient is established combined with the real-time residual signal. In the double-path filtering structure, the state prediction path uses the state sparse features for Kalman prediction, and the disturbance compensation path estimates the disturbance based on the residual and performs fusion. The system then adjusts the filtering gain and disturbance modeling parameters in conjunction with the error feedback, and finally outputs the estimated state and drives the anomaly detection module.

[0143] To fully evaluate the performance of the system, we choose a vertical test of a medium-sized liquid carrier rocket in October 2024 as a detailed test sample. During the test, the total time of the rocket vertical test is 540 seconds, the vertical angle is from horizontal 0° to vertical 90°, and during the 120th to 160th second of the vertical test, the platform encounters a 5-level side wave disturbance, causing the structural response to appear obvious disturbance. The traditional state estimation system fails to identify abnormal trends in time and only discovers deviations after the vertical test is completed, while the system of the present invention detects attitude estimation residual abnormalities at the 128th second, triggering state early warning in real time and outputting a judgment signal about 370 seconds in advance. The following table shows the performance comparison of the present invention method and the traditional method in three sea rocket vertical tasks:

[0144] Table 1 Comparison of state monitoring performance based on the present invention method and the traditional method

[0145]

[0146]

[0147] From the data results, it can be seen that the system of the present invention improves the response time of abnormal detection by nearly 6.6 times compared to the traditional method, effectively avoiding the risk of delayed response. In terms of state estimation accuracy, the maximum angle error is reduced from 3.8° to 1.1°, and the root mean square error is reduced by about 65%. At the same time, the false positive rate and the false negative rate are significantly reduced, improving the overall reliability and stability of the system.

[0148] In specific scenarios, the system has successfully coped with a low-temperature and high-humidity sea condition test in November 2024. In this task, due to the temporary wetting of the sensor interface, some strain data was lost. The system of the present invention completes the reconstruction of the lost data through a cooperative compressive sensing mechanism and maintains continuous and stable state estimation, avoiding misjudgment of the rocket control system on the vertical state and ensuring the safe completion of the task.

[0149] In addition, during the operation of the system, the residual driving linkage adjustment mechanism designed by the application can adaptively adjust the filtering and modeling parameters in multiple disturbance periods without manual intervention. According to the joint error index, the adjustment strength of the state prediction path and the disturbance compensation path can be adjusted independently, so that the filtering error converges rapidly to below 0.5° within 120 seconds, and the stable estimation ability is maintained in the subsequent stage, fully showing the effectiveness of the closed-loop feedback and dynamic adjustment structure.

[0150] In summary, the embodiment shows the robustness and precision performance of the method of the application in a complex disturbance environment. In particular, in the sea platform erecting state perception scene, it can respond to sudden disturbances, information missing and state abnormalities in real time, and construct a complete data perception-state prediction-disturbance compensation-feedback adjustment-anomaly identification closed loop process, providing a stable, efficient and deployable state monitoring technical solution for sea space launch erecting tasks.

[0151] The above is only the preferred specific embodiment of the application, but the protection scope of the application is not limited thereto. Any person skilled in the art can make equivalent replacements or changes within the technical scope disclosed by the application according to the technical solution and inventive concept of the application, which should be covered within the protection scope of the application.

Claims

1. A method for anomaly detection of sea-based rocket erecting state based on sparse representation and adaptive filtering, characterized in that, The method comprises the following steps: S1, acquiring original state observation data collected by multiple state monitoring sensors during the erecting process of the sea rocket, performing time alignment processing, and generating a multi-channel state observation time sequence; S2, constructing a dynamic sliding window mechanism according to the multi-channel state observation time sequence, setting an adaptive window length for each dynamic sliding window, and generating a dynamic sparse representation dictionary; S3, performing sparse coding on the state observation data in the dynamic sliding window based on the dynamic sparse representation dictionary, extracting a state sparse feature vector, and calculating a current state prediction value according to the state sparse feature vector; S4, calculating a residual signal between the current state prediction value and an actual state observation value, and calculating a feedback adjustment coefficient according to the residual signal; S5, constructing a double-path filtering architecture comprising a state prediction path and a disturbance compensation path, the state prediction path receiving the state sparse feature vector for performing state estimation, the disturbance compensation path receiving the residual signal for performing disturbance estimation, and fusing the outputs of the double paths to generate a final state estimation result; S6, according to the feedback adjustment coefficient, adjusting the filtering gain parameter used in the state estimation and the disturbance modeling parameter used in the disturbance estimation in linkage, and constructing a parameter adaptive adjustment mechanism; S7, inputting the final state estimation result and the residual signal into an anomaly detection model, and the anomaly detection model identifying state anomalies in the erecting process of the sea rocket based on a time sequence consistency analysis method and outputting an anomaly labeling result.

2. The sparse representation and adaptive filtering based anomaly detection method for sea launched rocket erect state according to claim 1, characterized in that, The S2 specifically comprises: S21, performing unified time axis mapping on the multi-channel state observation time sequence, and constructing a multi-channel observation matrix covering synchronous data of each channel; S22, setting an initial position of a sliding window in the multi-channel observation matrix, and extracting state observation data of each channel in a corresponding time period as a data input segment corresponding to the current sliding window; S23, for the data input segment corresponding to the current sliding window, calculating a state change rate of each channel in the time period, and constructing a nonlinear window length adjustment function according to the average value of the state change rates of all channels, defining the current window length L(t) as: wherein L min is the minimum length of the sliding window, L max is the maximum length of the sliding window, v i (t) is the rate of change of state of the i-th channel in the current time period, and N is the total number of state monitoring channels. S24, calculating a sliding residual signal between the state prediction value and the actual state observation value in a sliding period of the current sliding window, calculating a change amplitude Δε of the sliding residual signal, if the change amplitude Δε is greater than a set residual change threshold θ, moving the starting position of the sliding window by a sliding step Δt, updating the sliding window position, if Δε is less than or equal to the residual change threshold θ, keeping the starting position of the sliding window unchanged; S25, extracting an observation data segment in each sliding window triggered to slide, and constructing a set of dynamic sparse representation dictionaries corresponding to the time characteristics of each sliding window.

3. The sparse representation and adaptive filtering based anomaly detection method for sea launched rocket erect state of claim 1, wherein, The S3 specifically comprises: S31, extracting a state observation data segment in a time period corresponding to each sliding window, and constructing a to-be-encoded data matrix; S32, input the to-be-encoded data matrix as input, combine the dynamic sparse representation dictionary corresponding to the sliding window, and use the orthogonal matching pursuit algorithm to perform sparse coding on each state observation data to solve the sparse coefficient vector a, satisfying the following expression: x≈D·a, wherein x is a state observation vector, D is a dynamic sparse representation dictionary, a is a sparse coding coefficient vector, and a state sparse feature vector is extracted; S33, input the sparse coding coefficient vector a into the state prediction structure, which is composed of two nonlinear mapping functions, including a gating activation function and a state projection function: where, is the current state prediction value, W g , W p are the weight matrices for the gating path and the prediction path, respectively, b g , b p are the corresponding bias terms, and σ(·) is a nonlinear activation function. S34, arrange the state prediction values generated by the continuous sliding window in chronological order to construct a state prediction sequence.

4. The sparse representation and adaptive filtering based anomaly detection method for sea launched rocket erect state of claim 1, wherein, The S4 specifically comprises: S41, obtaining a current state prediction value in the constructed state prediction sequence and an actual state observation value y at the corresponding time t , respectively as the current prediction value and the current observation value; S42, predict value according to current state between the actual state observation value y t and the difference value calculates the residual signal ε t ; S43, based on a residual signal sequence {ε t-n ,…,ε t} of continuous multiple time points, calculate a residual change rate Δε t , which is the difference between the residual signal of the current time point and the residual signal of the previous time point. S44、According to the residual signal ε t With the residual change rate Δε t , the feedback adjustment function f(·) is constructed, and the feedback adjustment coefficient γ t , the feedback adjustment function adopts an exponential weight combination form.

5. The sparse representation and adaptive filtering based anomaly detection method for sea launched rocket erect state of claim 1, wherein, The S5 specifically comprises: S51, receive the state sparse feature vector, take the state sparse feature vector as the input of the state prediction path, and generate a state estimation value by using a prediction model in a Kalman filter structure for state estimation S52, receive the residual signal, take the residual signal as the input of the disturbance compensation path, and generate a disturbance estimation value d by using an exponential weighted residual modeling method for disturbance estimation t ; S53, weight and fuse the state estimation value generated by the state prediction path and the disturbance estimation value generated by the disturbance compensation path to construct a final state estimation value, and the fusion method uses a weighted average model.

6. The sparse representation and adaptive filtering based anomaly detection method for sea launched rocket erect state of claim 1, wherein, The S6 specifically comprises: S61, receive a feedback adjustment coefficient, and input the feedback adjustment coefficient into the state prediction path and the disturbance compensation path respectively to jointly adjust the filter gain parameter in the state estimation path and the disturbance modeling parameter in the disturbance compensation path; S62, in the state prediction path, dynamically update the filter gain parameter according to the value of the current feedback adjustment coefficient, and the filter gain parameter is used for the state update step in the Kalman filter structure; S63, in the disturbance compensation path, adjust the disturbance modeling parameter according to the change trend of the current feedback adjustment coefficient, and the disturbance modeling parameter includes the time weighting window length of disturbance estimation and the weighted distribution structure of historical residual signals; S64, in the current sliding window period, calculate the state estimation error between the state estimation value output by the state prediction path and the actual state observation value, and the disturbance estimation error between the disturbance estimation value output by the disturbance compensation path and the actual state observation value, and construct a joint error evaluation index; S65, according to the output result of the joint error evaluation index, dynamically adjust the adjustment step and update direction of the filter gain parameter used in the state prediction path and the disturbance modeling parameter used in the disturbance compensation path, and construct a dual-factor parameter adaptive adjustment mechanism coupled with linkage; S66, apply the filter gain parameter and the disturbance modeling parameter updated by the joint error to the execution process of the state prediction path and the disturbance compensation path at the next time, forming a continuous parameter self-optimization closed-loop system driven by the feedback adjustment coefficient as the core.

7. The sparse representation and adaptive filtering based anomaly detection method for sea launched rocket erect state of claim 6, wherein, The S65 specifically comprises: S651, based on the calculated state estimation error and disturbance estimation error, generate a joint error evaluation index, which is composed of the weighted combination of the state estimation error and the disturbance estimation error; S652, compare the joint error evaluation index with a preset joint error threshold, and determine whether to start the parameter adjustment process according to the comparison result, if the joint error evaluation index is greater than the preset joint error threshold, the parameter adjustment is performed, and if the condition is not met, the current parameter setting is maintained; S653. In parameter adjustment, the change value of the state estimation error is called as the adjustment control input to the state prediction path to adjust the filter gain parameter. The adjustment operation sets the adjustment step size and direction according to the error change value and updates the filter gain parameter. S654. Synchronously call the change value of the disturbance estimation error as a control input to the disturbance compensation path to adjust the disturbance modeling parameters. The disturbance modeling parameter adjustment strategy includes adjusting the time weighted distribution of the residual signal and the estimation window span. S655. Perform joint analysis on the error change value of the state prediction path and the error change value of the disturbance compensation path, and calculate the joint adjustment weight factor. The joint adjustment weight factor is used to allocate adjustment resources between the two paths. S656 will output the updated filter gain parameters and disturbance modeling parameters respectively, and pass them to S66 to build a continuous dual-path parameter linkage update chain.

Images