Ship lock operation state remote monitoring system based on Internet of Things

Abstract

The invention relates to the technical field of Internet of Things and ship lock management, and discloses a ship lock operation state remote monitoring system based on Internet of Things, which comprises a data acquisition module used for acquiring multi-source sensing data; the edge preprocessing module is used for carrying out compression, denoising and format unification processing on the multi-source sensing data; the space-time alignment and fusion module is used for carrying out alignment registration based on the timestamp and the space pose information of the processed multi-source data so as to construct a multi-dimensional fusion tensor, and the tensor modeling module introduces geometric constraint information obtained by Lie group space registration on the basis of the fusion tensor so as to construct the multi-dimensional fusion tensor. And a tensor optimization model with spatial structure constraints is constructed, and key tensor features are extracted. According to the method, the technical scheme of'multi-source data fusion, Lie group registration and tensor structure modeling 'is adopted, and the technical effect of unified modeling and spatial registration of the multi-dimensional operation state of the ship lock is achieved.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things technology and ship lock management technology, and in particular to a ship lock operation status remote monitoring system based on the Internet of Things. Background Art

[0002] As an important facility for water transportation, the operating status of locks directly affects the navigation safety and efficiency of ships. With the rapid development of Internet of Things technology, applying Internet of Things technology to the field of lock management to realize real-time monitoring and remote management of lock operating status has become a hot topic of current research.

[0003] However, in traditional ship lock operation monitoring systems, independent sensors are mostly deployed at fixed points, and electrical parameters or hydraulic signals are usually used as the main basis for judgment. Although this type of method has achieved basic state perception, it is often unable to cope with the problem of collaborative fusion of multi-source heterogeneous data. The sampling frequency, clock accuracy and data structure of different sensors are inconsistent, which makes it difficult to synchronize data in the time and space dimensions, the fusion cost is high, and the real-time performance is poor. The currently commonly used remote communication mechanism has shortcomings in data reliability. Most solutions rely on a stable network to upload status data. Once the communication link is damaged or interrupted, the system will lose the ability to track key operating states. Especially in emergencies, fault-tolerant reconstruction is impossible, and data is lost. Some systems Although the system has introduced a cache mechanism, it still cannot effectively interpolate missing information. In terms of state modeling, existing methods mostly use traditional linear interpolation, Kalman filtering and other algorithms for prediction or estimation. Such methods are barely applicable under short-term and stable working conditions. However, when facing a dynamic system such as a lock that is strongly affected by hydrological and structural coupling, the linear model is obviously insufficient. It lacks the processing of spatial information such as rotation and posture, has low prediction accuracy and poor stability, and the error accumulation is obvious in long-term use. In addition, although digital twin technology has been gradually introduced into lock monitoring, most implementations are still in the "static modeling" stage and it is difficult to make adaptive adjustments as the system's operating status continues to change. Once the model deviates from reality, it will continue to output misleading prediction information. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a remote monitoring system for the operation status of a ship lock based on the Internet of Things, which solves the problems of the inability to collaborate with heterogeneous data, fragmented state perception, untraceable status and broken monitoring chain when the network is interrupted, model failure and decreased accuracy when actual operation changes, insufficient dimensions in dynamic fault characterization, and missing structural information.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A remote monitoring system for ship lock operation status based on the Internet of Things, comprising the following steps:

[0006] Data acquisition module, used to obtain multi-source perception data;

[0007] Edge preprocessing module, used to compress, denoise and unify the format of multi-source perception data;

[0008] The spatiotemporal alignment and fusion module is used to align and register the processed multi-source data based on their timestamps and spatial pose information, thereby constructing a multi-dimensional fusion tensor.

[0009] The tensor modeling module, based on the fusion tensor, introduces the geometric constraint information obtained by Lie group space registration, and then constructs a tensor optimization model with spatial structure constraints to extract key tensor features;

[0010] Fault prediction module, which is used to build a fault propagation path model based on key tensor features and generate predicted status information;

[0011] The digital twin verification module is used to build a virtual twin system based on predicted state information and compare its operating state with the actual collected data in real time. The prediction model is adaptively corrected through an error feedback mechanism to form a dynamic closed-loop verification mechanism, while also generating key state data for remote synchronization.

[0012] The communication and fault-tolerance module is used to receive key status data and transmit it remotely through a data communication protocol that includes verification parameters generated based on Lie group interpolation. It also interpolates and reconstructs missing information and maintains synchronization in the event of network interruption or data loss.

[0013] Preferably, the data acquisition module includes:

[0014] Vibration acquisition unit, used to obtain vibration signals of gates or opening and closing equipment;

[0015] Water level detection unit, used to obtain water level change data upstream and downstream of the lock;

[0016] Hydraulic signal acquisition unit, used to collect the working pressure and flow parameters of the opening and closing hydraulic system;

[0017] Electrical parameter acquisition unit, used to detect voltage, current and power data of the control circuit;

[0018] The image acquisition unit is used to collect image information of the lock operation area in real time.

[0019] Preferably, the edge preprocessing module includes:

[0020] Format unification unit, used to convert sensor data of different sources and types into a standard time series format;

[0021] A data compression unit, used for compressing and encoding multi-source sensing data to reduce transmission bandwidth;

[0022] The noise suppression unit is used to remove signal noise based on statistical feature filtering or wavelet transform.

[0023] Preferably, the spatiotemporal alignment and fusion module includes:

[0024] The time alignment unit is used to interpolate and align multi-source data according to their timestamps;

[0025] The spatial registration unit is used to uniformly map various sensor data to a reference coordinate system using the Lie group registration method based on the spatial posture information of each sensor device;

[0026] The tensor construction unit is used to fuse the aligned multi-source data into a multi-dimensional tensor structure.

[0027] Preferably, the tensor modeling module includes:

[0028] A geometric constraint extraction unit, used to extract geometric structure information based on Lie group pose relationships;

[0029] Spatial structure modeling unit, used to embed geometric constraint information into tensor optimization models;

[0030] The feature extraction unit is used to extract key low-dimensional features that characterize the operating status of the lock through a tensor decomposition algorithm.

[0031] Preferably, the fault prediction module includes:

[0032] A state propagation modeling unit, used to build a fault propagation path model based on key tensor features;

[0033] A state deduction unit is used to deduce the evolution trend of potential system failures based on the constructed path model;

[0034] The prediction output unit is used to output the prediction status information for use by subsequent modules.

[0035] Preferably, the digital twin verification module includes:

[0036] A virtual modeling unit, used to construct a digital model of a virtual ship lock based on the predicted state information;

[0037] A state comparison unit is used to synchronously compare the virtual model's operating state with the actual collected data;

[0038] Model adaptive correction unit, used to update and adaptively adjust the parameters of the prediction model based on the error feedback mechanism;

[0039] The state synchronization generation unit is used to generate key state data consistent with the actual state for remote transmission.

[0040] Preferably, the communication and fault tolerance module includes:

[0041] A check parameter generating unit, configured to generate redundant check parameters for state data based on a Lie group interpolation method;

[0042] A data communication unit for remote information transmission according to a data communication protocol embedded with verification parameters;

[0043] The fault-tolerant reconstruction unit is used to reconstruct missing data based on Lie group interpolation in the event of data interruption or loss.

[0044] Preferably, the tensor modeling module adopts a tensor optimization model as follows:

[0045] Let the fused tensor be By introducing the geometric constraint matrix G∈R I×R , construct the following optimization objective function:

[0046]

[0047] in, Represents the original fusion tensor, which contains multi-source perception data of I, J, and K dimensions, usually corresponding to sensor type, time series, space or feature dimension, G∈R I×R Represents the introduced geometric constraint matrix, which is used to impose a priori constraints on the first factor matrix A in the tensor decomposition, reflecting the spatial structure or posture correlation. R represents the rank of the tensor, which is equivalent to the rank number or number of components of the decomposition, and is the number of low-dimensional vectors used to approximate the tensor. r ∈R I 、b r ∈R J 、c r ∈R K are the decomposition vectors of the rth tensor rank component on three modules, which are used to form a rank tensor, represents the outer product operation of vectors, that is is a rank tensor, Represents the reconstructed tensor expression of the CP decomposition of the tensor, A∈R I ×R Indicates that all a r The matrix after the vector column combination, that is, A=[a1,a2,...,a R ], is the factor matrix on the first module, Represents the square of the Frobenius norm, that is, the sum of the squares of all elements of a tensor or matrix, which is used to measure the reconstruction error or matrix deviation, λ∈R + is the regularization coefficient, which is used to balance the influence between the main reconstruction error term and the geometric constraint term.

[0048] Preferably, the fault-tolerant reconstruction unit in the communication and fault-tolerant module estimates the missing data using a Lie group interpolation method, and the interpolation formula is:

[0049] Assume that the two adjacent observation values ​​of the key state data on the Lie group are g1, g2∈SE(3), and the time corresponding to the missing state is t∈(t1, t2), then the estimated state is:

[0050]

[0051] Among them, g1,g2∈SE(3) represent two known Lie group observations at time points t1 and t2, describing the spatial pose information, t∈(t1,t2) represents the target time point to be interpolated, which is located between the two observations, and g(t) represents the Lie group element estimated at time t, that is, the interpolation result of the target state on SE(3). represents the inverse matrix of g1, representing the transformation from the reference coordinate system at time t1 to the global or target coordinate system, represents the relative motion from g1 to g2, log(·) represents the logarithmic mapping from Lie group to Lie algebra, Returns the vector or corresponding matrix of 6-dimensional motion. exp(·) represents the exponential mapping from Lie algebra to Lie group. Used to recover the rigid body transformation from the velocity field, Represents the normalized interpolation weight of the current time t relative to the two observation moments, represents the “motion difference” from g1 to g2 in Lie algebra, and exp(·) represents the scaling of the motion difference and its mapping back to SE(3) space to achieve interpolation.

[0052] The present invention provides a remote monitoring system for ship lock operation status based on the Internet of Things.

[0053] Beneficial effects:

[0054] 1. The present invention adopts the technical solution of "multi-source data fusion, Lie group registration and tensor structure modeling" to achieve the technical effect of unified modeling and spatial registration of the multi-dimensional operating status of the lock. Compared with the technical solution in the existing technology that is only based on independent monitoring of single-source sensors, it solves the shortcomings of the existing technology in terms of the inability to coordinate heterogeneous data and fragmented state perception.

[0055] 2. The present invention adopts a data communication protocol technology solution based on the fusion of Lie group interpolation and verification parameters, which achieves the technical effect of realizing state self-recovery and fault-tolerant transmission in the case of communication anomalies or data loss. Compared with the technical solution in the existing technology that relies on complete data link transmission, it solves the shortcomings of the state being untraceable and the monitoring chain being broken when the network is interrupted.

[0056] 3. The present invention adopts a closed-loop error feedback verification mechanism driven by digital twins to achieve the technical effect of adaptively correcting and dynamically updating the remote prediction state. Compared with the technical solution in the existing technology that the static prediction model does not have the ability to make online corrections, it solves the shortcomings of model failure and decreased accuracy when facing actual operation changes.

[0057] 4. The present invention adopts a tensor optimization modeling technology solution that integrates geometric constraints to achieve high-fidelity extraction of key low-rank features in high-dimensional perception data. Compared with the technical solution that relies on planar data compression in the existing technology, it solves the shortcomings of insufficient dimensionality and missing structural information in dynamic fault characterization. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of the system architecture of the present invention;

[0059] Figure 2 Schematic diagram of the data acquisition module of the present invention;

[0060] Figure 3 Schematic diagram of the edge preprocessing module of the present invention;

[0061] Figure 4 Schematic diagram of the spatiotemporal alignment and fusion module of the present invention;

[0062] Figure 5 This is a schematic diagram of the tensor modeling module of the present invention;

[0063] Figure 6 This is a schematic diagram of a fault prediction module of the present invention;

[0064] Figure 7 This is a schematic diagram of the digital twin verification module of the present invention;

[0065] Figure 8 Schematic diagram of the communication and fault-tolerant module of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] Please see the attached Figure 1-8 , an embodiment of the present invention provides a remote monitoring system for the operation status of a ship lock based on the Internet of Things, including a data acquisition module, an edge preprocessing module, a spatiotemporal alignment and fusion module, a tensor modeling module, a fault prediction module, a digital twin verification module, and a communication and fault tolerance module.

[0068] The following is a detailed description of each module in the system of the present invention, which comprehensively explains the specific implementation principles, technical details and processes of each module.

[0069] Data acquisition module

[0070] The data acquisition module of the present invention includes a vibration acquisition unit, a hydraulic signal acquisition unit, an electrical parameter acquisition unit, and an image acquisition unit. The specific implementation is as follows:

[0071] The vibration acquisition unit is equipped with a three-axis accelerometer and a high-speed data acquisition chip on the main shaft of the gate hoist, the gate trunnion and the hanging point to capture the tiny vibration fluctuations generated during the operation of the machine. The signal is recorded in the form of a three-dimensional acceleration time series.

[0072]

[0073] Among them, a x (t),a y (t),a z (t): acceleration in three directions corresponding to the spatial rectangular coordinate system at time t, a(t): vibration feature vector at time t, which will be used in subsequent fault feature extraction and spectrum analysis steps.

[0074] The water level detection unit uses a non-contact radar-type ranging water level meter deployed in the upstream and downstream pilot channels to collect water level height per unit time. The water level data is processed in the following form:

[0075] Upstream water level: H u (t), in meters;

[0076] Downstream water level: H d (t), in meters;

[0077] Water level difference: ΔH(t) = H u (t)-H d (t), which is used to judge the gate opening and closing conditions and hydraulic stability.

[0078] Water level data is collected at a frequency of 5Hz, with a sampling accuracy better than ±1cm. It has a drift correction algorithm module to avoid false triggering due to rain and wave interference.

[0079] The hydraulic signal acquisition unit is used to collect the pressure and flow of each branch of the opening and closing hydraulic system. The unit is equipped with pressure sensors P(t) and flow sensors Q(t) at the output of the hydraulic pump group, the valve group interface, and the inlet and outlet of the cylinder. The collected signals are used to calculate the instantaneous hydraulic power and evaluate the loading status of the actuator. The hydraulic power calculation relationship is:

[0080] Wh (t) = P(t)·Q(t);

[0081] Where P(t) represents the actual pressure value of the hydraulic system at time t, in Pa; Q(t) represents the hydraulic flow value, in cubic meters per second, W h (t) represents the mechanical power output of the hydraulic system at time t, in watts.

[0082] This power signal will be used as part of the execution force feature vector in the subsequent tensor modeling module for feature fusion. The electrical parameter acquisition unit is set between the control power distribution box and the main contactor. The Hall sensor and the power metering unit collect electrical parameters such as voltage U(t), current I(t), and power factor cosφ(t) in real time. The system power is estimated using the following relationship:

[0083] P e (t) = U(t)·I(t)·cosφ(t);

[0084] Among them, U(t) represents the voltage signal at the sampling moment, the unit is volt, I(t) represents the current value at the sampling moment, the unit is ampere, cosφ(t) represents the electrical power factor, reflecting the nature of the load, P e (t) represents active power in watts.

[0085] The sampling period is set to 1 second per time, and the sliding window method is used to filter out short-term spikes to improve the robustness of the input signal.

[0086] The image acquisition unit is used to collect images of the lock operation area and the operating status of key components. The unit deploys several high-definition industrial cameras and supports night vision supplementary lighting and rain and fog enhancement functions. The image data is encapsulated in the form of time-stamped frames, which are expressed as follows:

[0087]

[0088] Where t represents the acquisition time corresponding to the image frame, and x and y represent the pixel coordinates in the image; Represents the grayscale or color information of the image at time t. If it is a color image, it is a three-channel tensor.

[0089] Image frames are uploaded at a fixed resolution, and structured features are extracted by the edge processing module and fused into the state tensor.

[0090] Each data acquisition unit encapsulates data through a unified interface protocol in the form of a triple structure:

[0091] D i ={T i ,L i ,V i};

[0092] Among them, T i Indicates the sampling timestamp, with millisecond accuracy, L i Indicates the equipment number and installation posture code, V i Represents the perception data value of the corresponding unit, which may be in scalar, vector or tensor form.

[0093] All acquisition units are uploaded to the edge node via a wired bus (CAN, RS485) or wireless protocol (LoRa, ZigBee). The node has a built-in unified parser that pre-checks, frames, and parses the original data packets to ensure the temporal and spatial consistency of multi-source signals.

[0094] To ensure stable operation of the acquisition system, each unit features breakpoint resume, data buffering, and power-off self-recovery. Inter-node clock synchronization is based on GPS time signals, ensuring cross-module data timing alignment accuracy better than 1 millisecond.

[0095] Edge preprocessing module

[0096] The edge preprocessing module of the present invention includes a format unification unit, a data compression unit, and a noise suppression unit. The specific implementation is as follows:

[0097] The format unification unit receives raw data streams from different sampling frequencies and different data structures, first performs timestamp correction and synchronization, then uses the resampling algorithm to unify the data time interval and outputs standardized time series data.

[0098] Specifically, let the original sampling sequence be

[0099] D i ={(T i,k ,V i,k )|k=1,...,K i};

[0100] Among them, T i,k represents the timestamp of the kth sampling of the i-th sensor, V i,k Represents the corresponding sample value, which may be a scalar, vector or tensor.

[0101] The format unified unit constructs an equally spaced time series based on a unified sampling period Δt:

[0102] S i ={(t j ,s j )|t j =t0+jΔt,j=0,1,...,N};

[0103] Among them, t0 is the unified starting time point, s j Through the interpolation function fi (·)calculate:

[0104] s j =f i (t j )≈V i,k Interpolation estimate

[0105] The interpolation function is preferably based on cubic spline interpolation, and the formula is:

[0106]

[0107] Among them, f i (t) represents the function f i The value at time point t. This function is usually used to describe the system state at a certain time t. It may be the interpolation or approximation of a variable over a period of time. represents the summation of exponents m from 0 to 3, which indicates that the function f i (t) is a cubic polynomial, that is, it contains at most the third power of t, a m Represents the coefficients in the polynomial, specifically the coefficients corresponding to the mth power, (tt k ) m Represents the time point t and a reference time point t k The difference between the mth power, this difference is used to construct the various items of the function to determine the impact of different time points on the function value, t k Represents a reference time point, usually a discrete time point, used to calculate the difference between t and it, T ik Indicates the starting time of the time interval, usually the beginning of the i-th time period, and T i,k+1 Together they constitute a time period, T i,k+1 Indicates the end time of the time interval, usually the end of the i-th time period, used to limit f i (t) defines the time range, t∈[T ik ,T i,k+1 ] represents the value range of t, that is, t must be in the interval [T ik ,T i,k+1 ], which usually indicates that the function is effective in this time interval, where the coefficient a m According to the adjacent sampling point V i,k and boundary conditions are calculated.

[0108] Through this method, the format unification unit realizes the time domain alignment and format unification of multi-source sensor data, providing complete and continuous data input in time sequence for subsequent processing modules.

[0109] The data compression unit uses the Wavelet Packet Transform (WPT) algorithm to perform time-frequency domain compression on the sequence after format unification. i , the compression steps are as follows: First, the sequence S i Perform multi-layer wavelet packet decomposition, set the decomposition layer number to J, and obtain the frequency band sub-signal:

[0110] W j,k =∑ n s n ·ψ j,k (n);

[0111] Among them, W j,k is the coefficient of the kth frequency band of the jth layer, ψ j,k (n) is the corresponding wavelet packet basis function, s n For sequence S i The nth sample value in .

[0112] Then, based on the energy compression principle, the energy of each frequency band is calculated:

[0113] E j,k =∑ n |W j,k (n)| 2 ;

[0114] Among them, E j,k Represents the energy value at the jth row and kth column position. This value is usually used to describe the intensity, amplitude or power of certain signals or data points. It often appears in signal processing or image processing. j,k (n)| 2 W j,k (n) The square of the absolute value at position n, where W j,k (n) is usually a complex value (for example in wavelet transform, Fourier transform or filtering process), |W j,k (n)| is its magnitude, and the square represents the energy or power.

[0115] The frequency band coefficients whose cumulative energy ratio reaches a threshold value η (such as 95%) are selected and retained, and the other coefficients are set to zero, thereby achieving sparse representation. Finally, the compressed signal is reconstructed by inverse wavelet packet transform, which is expressed as:

[0116]

[0117] in, are the coefficients retained after thresholding.

[0118] The noise suppression unit combines statistical filtering and wavelet denoising methods to further improve signal quality.

[0119] Statistical filtering includes:

[0120] Mean filtering:

[0121]

[0122] in, It represents the value after smoothing the j-th position in the original sequence, which is usually used for denoising or interpolation. It is the estimated result after local averaging. j+m Represents the observation value at the j+mth position in the original sequence, where m is an offset ranging from -M to +M, representing a window of values ​​around the jth position. Indicates S j As the center, the sum of the M sampling points before and after it is made to form a sliding sum of the local window It is a normalization factor, which indicates the total number of samples contained in the sliding window. The window size is 2M+1, that is, M points to the left and right of the current point. j indicates the index position of the current smoothing, which is usually the position index of a discrete time series or spatial sequence.

[0123] Median filter:

[0124]

[0125] in, It represents the result of applying median filtering to the jth position in the original sequence, which is the smoothed estimate of the position. median{·} means taking the median of all the values ​​in the brackets, that is, taking the middle value after sorting the values. Median filtering is highly robust to outliers (such as spike noise). j-M ,S j ,…,S j+M Indicates that the current point S j The sliding window data set is centered on M adjacent values ​​taken forward and backward, with a total of 2M+1 values. j represents the index of the currently filtered data point in the sequence, representing the current processing position. M represents the half-width of the sliding window, which determines the context range involved in the median calculation. The total window size is 2M+1.

[0126] Wavelet denoising uses a soft threshold function:

[0127]

[0128] in, It represents the output result after applying the soft threshold operation to the input w. It is a nonlinear transformation and is often used in sparse representation and signal denoising. w represents the input variable, which is usually a coefficient, residual value, gradient component or element in the transform domain (such as wavelet domain or Fourier domain) in the signal. λ represents the threshold parameter, which controls the intensity of the soft threshold operation. A larger λ will suppress more small-amplitude components, resulting in a stronger sparse effect or denoising ability. |w| represents the absolute value of w, which is used to determine whether it is greater than the set threshold λ. sign(w): represents the sign function, which is defined as follows:

[0129] If w>0, then sign(w)=+1;

[0130] If w<0, then sign(w)=-1;

[0131] If w=0, then sign(w)=0;

[0132]

[0133] Where σ is the standard deviation of the noise estimate and N is the data length.

[0134] Through multi-scale wavelet decomposition and coefficient threshold processing, high-frequency random noise can be effectively suppressed to ensure the preservation of signal characteristics.

[0135] In this edge preprocessing module, the format unification unit first synchronizes and converts the time format of multi-source data, outputting a standard time series. The noise suppression unit filters and denoises the unified series, and the data compression unit then performs time-frequency conversion and compression encoding on the purified signal. The module utilizes pipeline processing to ensure high real-time performance and stability.

[0136] Spatiotemporal alignment and fusion module

[0137] The spatiotemporal alignment and fusion module of the present invention includes a time alignment unit, a spatial registration unit, and a tensor construction unit. The specific implementation is as follows:

[0138] The spatiotemporal alignment and fusion module includes a time alignment unit. In order to solve the problem of inconsistent sampling times of different sensors, the time alignment unit uses an interpolation algorithm to synchronize the unified time series. Specifically, let the sampling sequence of the i-th sensor be

[0139] S i ={(t i,k ,v i,k )|k=1,2,…,K i};

[0140] Among them, t i,k is the timestamp of the kth sampling of the i-th sensor, v i,k is the corresponding sampling value. The unified reference time series is

[0141] T={t j |j=1,2,…,N};

[0142] The time alignment unit uses a piecewise cubic Hermite interpolation method to ensure that the interpolated signal is smooth and has no oscillation. The interpolation formula is:

[0143]

[0144] in, is sensor i at the same time point t j The interpolation estimate of t i,k Indicates the timestamp of the kth sampling of the i-th sensor, in seconds (s), v i,k Represents the corresponding sample value, which depends on the sensor type and can be a scalar or a multidimensional vector, t j It represents the jth time point in the unified time series, in seconds (s), and N represents the length of the unified time series.

[0145] The spatial registration unit uses the Lie group SE (3) registration method to achieve unified mapping of spatial data based on the spatial posture information of each sensing device. Let the spatial posture of the i-th sensor relative to the reference coordinate system be

[0146]

[0147] Among them, R i ∈SO(3) is the rotation matrix, which describes the rotation relationship of the sensor coordinate system around the reference coordinate system. is the translation vector, describing the position of the origin of the sensor coordinate system in the reference coordinate system,

[0148] The spatial point or data vector p collected by the sensor i ∈R 3 Mapping to the reference coordinate system is represented as

[0149]

[0150] Among them, p i Represents the position vector of point i in the local coordinate system, usually a column vector in three-dimensional space, in the form of [x, y, z] T , Indicates that point p i After transformation g i After that, it is mapped to the position in the reference coordinate system, that is, the expression under the global coordinate or reference coordinate, R i represents the rotation matrix of the i-th coordinate system, which belongs to the Lie group SO(3) and is used to map the direction in the local coordinate system to the global coordinate system t iRepresents the translation vector of the i-th coordinate system in the global reference system, belonging to R 3 , represents the position of the origin of the coordinate system, g i represents the i-th rigid body transformation or posture transformation, which belongs to the Lie group SE(3) and is composed of the rotation matrix R i and the translation vector t i The combined affine transformation acts on point p i The rotation and translation operations are completed at the same time.

[0151] The tensor construction unit constructs a multi-dimensional tensor based on the data after spatiotemporal alignment and spatial registration, which is formally expressed as

[0152]

[0153] Where N is the number of tensor dimensions, usually including time dimension, space dimension and sensor channel dimension; I1,I2,…,I N are the sizes of each dimension, such as time series length, number of spatial grid points, number of sensor types, etc.

[0154] The specific construction process includes: first arranging the sequence output by the time alignment unit according to the time dimension, then mapping the spatial point data converted by the spatial registration unit to the spatial dimension index, and finally mapping the sensor type, acquisition features, etc. to the channel dimension.

[0155] Each element in the tensor

[0156]

[0157] Corresponding to the i1th time point, i2th spatial position and i N The fusion value of the feature channel.

[0158] This module adopts a pipeline approach, and the data flow passes through the time alignment unit, spatial registration unit, and tensor construction unit in sequence to achieve a unified time base, spatial base, and multi-dimensional fusion of the data.

[0159] The module has high concurrent processing capabilities and data consistency assurance, adapting to the complex multi-sensor environment of ship locks.

[0160] This spatiotemporal alignment and fusion module provides standard input format and accurate data support for subsequent tensor decomposition-based fault diagnosis and digital twin simulation, and has good engineering feasibility and scalability.

[0161] Tensor Modeling Module

[0162] The tensor modeling module of the present invention includes a geometric constraint extraction unit, a spatial structure modeling unit, and a feature extraction unit. The specific implementation is as follows:

[0163] The geometric constraint extraction unit is responsible for extracting spatial pose information from multi-source data. Specifically, based on the Lie group SE(3) spatial registration method, the relative pose transformation g between each sensor is extracted. ij ∈SE(3), by calculating the relative transformation of each pair of sensors, we can obtain the rotation matrix R between the sensors. ij ∈SO(3) and translation vector t ij ∈R 3 , thus providing geometric constraints for subsequent tensor modeling.

[0164] Assume that the relative pose transformation between the i-th sensor and the j-th sensor is:

[0165]

[0166] Among them, g i The i-th element of a system state or object on the Lie group is usually used to represent the posture of a rigid body, posture transformation, rotation matrix or transformation matrix, etc. It is an element belonging to the Lie group G, such as a point in SO(3) or SE(3), g j Represents the jth Lie group element, which has the same meaning as g i The same, representing the posture or state of another moment or another node, is a Lie group element g i Because the Lie group is a group structure, each element has a unique inverse, which represents the inverse from the state g i Back to the transformation of the unit element e, g ij Indicates that from g i to g j The relative transformation of , also known as the right-multiplied difference on the Lie group.

[0167] The spatial structure modeling unit uses geometric constraint information and embeds it into the tensor optimization model. This unit constructs the final optimization objective function by minimizing the tensor reconstruction error and combining spatial pose constraints. Let the fused tensor be Represents the fusion tensor of multi-source data, where I1, I2, ..., I N Represent the dimensions of time, space, and sensor features respectively. The spatial structure constraint is introduced by the geometric constraint matrix To adjust the final structure of the tensor. The optimization objective function is expressed as follows:

[0168]

[0169] in, Represents the original fusion tensor, which contains multi-source perception data of I, J, and K dimensions, usually corresponding to sensor type, time series, space or feature dimension, G∈R I×RRepresents the introduced geometric constraint matrix, which is used to impose a priori constraints on the first factor matrix A in the tensor decomposition to reflect the spatial structure or posture correlation. R represents the rank of the tensor (i.e., the rank in CP decomposition), which is equivalent to the rank number or the number of components of the decomposition. It is the number of low-dimensional vectors used to approximate the tensor. r ∈R I 、b r ∈R J 、c r ∈R K are the decomposition vectors of the rth tensor rank component on three modes, which are used to form a rank tensor. represents the outer product operation of vectors, that is is a rank tensor, Represents the reconstructed tensor representation of the CP (CANDECOMP / PARAFAC) decomposition of the tensor, A∈R I×R Indicates that all a r The matrix after the vector column combination, that is, A=[a1,a2,...,a R ], is the factor matrix on the first module, Represents the square of the Frobenius norm, that is, the sum of the squares of all elements of a tensor or matrix, which is used to measure the reconstruction error or matrix deviation, λ∈R + is the regularization coefficient (weight parameter) used to balance the influence between the main reconstruction error term and the geometric constraint term.

[0170] The feature extraction unit extracts the low-dimensional features of the lock operation status through the tensor decomposition algorithm, and decomposes the optimized tensor using the high-order singular value decomposition (HOSVD) or alternating least squares (ALS) method. Its decomposition can be expressed as:

[0171]

[0172] Among them, X * Represents the estimated tensor (or multidimensional array), which is usually a reconstructed or approximated version of the original tensor X. In the fault-tolerant model, X * It is the tensor result after interpolation or optimization of missing data. Represents a set of factor matrices or factor tensors, corresponding to each mode or dimension of the tensor. These factors reflect the potential structure or low-rank features of the data in different dimensions and are obtained through decomposition learning or optimization. Multiple modular products are performed in sequence to achieve dimension-by-dimension reconstruction of the entire tensor. N represents the order of the tensor (that is, the number of dimensions of the tensor).

[0173] The optimization objective function is:

[0174]

[0175] Among them, X * Indicates that this is the target tensor, representing the original multidimensional data that we want to approximate or reconstruct. Represents a set of factor matrices or factor tensors, corresponding to the feature representation of the original tensor in each dimension (also called mode). These factors are the variables that need to be solved through optimization. × represents the mode-n product operator between the tensor and the factor, which represents the tensor reconstruction process of step-by-step multiplication of each dimension with the corresponding factor. 2 It represents the square of the Frobenius norm, that is, the sum of the squares of the differences between all elements in the tensor, which is used to measure the error between the estimated result and the target tensor. N represents the order of the tensor, that is, the number of dimensions it contains. For example, a three-dimensional tensor (time × space × state variable) has N = 3. min means to minimize the objective function through optimization, that is, to minimize the difference between the estimated tensor and the target tensor.

[0176] By minimizing this objective function, key low-dimensional features can be extracted from high-dimensional tensors for analysis of lock operation status, fault diagnosis, performance evaluation, etc.

[0177] The specific process includes: first, the geometric constraint extraction unit extracts timestamps and spatial pose information from multi-source sensor data and calculates the relative pose transformation between sensors. Then, the spatial structure modeling unit constructs a tensor optimization model with spatial structure constraints based on the geometric constraint information. Then, the feature extraction unit uses the tensor decomposition algorithm to extract low-dimensional features from the optimized tensor, and finally obtains the key features for monitoring the operation status of the lock.

[0178] Fault prediction module

[0179] The fault prediction module of the present invention includes a state propagation modeling unit, a state deduction unit, and a prediction output unit. The specific implementation is as follows:

[0180] The state propagation modeling unit is used to build a fault propagation path model based on key tensor features. Key tensor features are low-dimensional features extracted from the previous module (such as the tensor modeling module) that characterize various aspects of the lock's operating status.

[0181] Using these features, the state propagation modeling unit constructs a propagation model based on the association between states to simulate the path of fault propagation in the system.

[0182] Assume that the propagation process of the fault state is described by the state transition matrix T, where each state s i Corresponding to a specific operating state or fault state in the system. The propagation path model of the fault state is based on the following equation:

[0183] s t+1 =T·s t ;

[0184] Among them, s t represents the fault state of the system at time t; T is the state transition matrix, which describes the transition probability from the current state to the next state; s t+1 is the fault state of the system at time t+1.

[0185] By analyzing historical data and key tensor features, the state propagation modeling unit can infer the propagation path of potential faults based on the actual operating status of the system, further providing a basis for subsequent fault prediction.

[0186] Next, the state deduction unit deduces the evolution trend of potential system faults based on the constructed fault propagation path model.

[0187] The state deduction unit uses the fault propagation path model to deduce the future state of the system and predict potential fault evolution trends.

[0188] During the deduction process, the current state of the system s t It is input into the deduction algorithm, and the algorithm calculates the state s at the future time t+k based on the state transition matrix T t+k , where k represents the step length of the deduction. The deduction process can be expressed as:

[0189] s t+k =T k ·s t ;

[0190] Among them, T k represents the kth power of the state transfer matrix T, which represents the fault propagation from the current time t to the future time t+k; s t+k Predicts the fault state at the future time t+k. This unit analyzes the evolution trend of potential faults through multi-step deduction and generates the future fault evolution path of the system, thereby providing support for system maintenance and fault warning.

[0191] This unit analyzes the evolution trend of potential faults through multi-step deduction, generates the system's future fault evolution path, and provides support for system maintenance and fault warning.

[0192] Finally, the prediction output unit outputs predicted state information for use by subsequent modules. This unit converts the fault prediction information generated by the state deduction unit into a specific predicted state for processing by the subsequent decision-making module. The output predicted state information includes the potential fault's occurrence time, fault location, and impact range.

[0193] The format of the prediction output is:

[0194]

[0195] in, is the predicted fault status information, t k For the predicted moment, At time t k Predicted system failure conditions.

[0196] The prediction output unit passes this information to subsequent modules for fault analysis, diagnosis or alarm.

[0197] Digital Twin Verification Module

[0198] The digital twin verification module of the present invention includes a virtual modeling unit, a state comparison unit, a model adaptive correction unit, and a state synchronization generation unit. The specific implementation is as follows:

[0199] The virtual modeling unit is used to construct a digital model of the virtual ship lock based on the predicted state information. The construction process of the virtual model is based on the actual working state of the system and the predicted state information.

[0200] By establishing a mapping between the physical model and the system dynamic model, the virtual modeling unit uses the predicted status information output from the previous module (such as the fault prediction module) to dynamically update the digital twin model of the virtual lock.

[0201] Assume that the digital model of the virtual lock is Where N and M are the spatial and temporal dimensions of the model respectively. The running state of the virtual model is updated by the following equation:

[0202]

[0203] in, represents the virtual model state predicted at time t+1, which is an estimate of the system's operating state at the next moment. It is determined by the current state and input drive. f(·) is the system dynamic update function, which describes how the virtual lock is updated based on the state of the previous moment and the predicted state information at the current time step. is the predicted status information at time t, including the current operating status of the lock, equipment health status and other influencing factors.

[0204] The state comparison unit is used to perform real-time synchronous comparison between the operating state of the virtual lock model and the actual collected sensor data. The actual collected data is transmitted to the unit through the edge computing device and compared with the operating state of the virtual model to detect the accuracy of the prediction model. Assume that the actual collected data is This data contains various sensor measurements of the lock, including temperature, pressure, flow, etc.

[0205] The state comparison unit determines the model's fitting effect by calculating the error between the virtual model and the actual data. The comparison process can be quantified using the following error formula:

[0206]

[0207] in, represents the error at time t, reflecting the difference between the virtual lock model and the actual data. is the Frobenius norm, which represents the square root of the sum of the squares of the matrix or tensor elements.

[0208] The state comparison unit calculates error feedback and tracks the deviation between the virtual model and the actual state in real time to ensure the accuracy of the virtual model.

[0209] The model adaptive correction unit adaptively corrects the virtual model based on the error feedback mechanism. Based on the error information output by the state comparison unit, the model adaptive correction unit dynamically adjusts the prediction parameters in the virtual model to improve the model's accuracy. The model adaptive correction unit updates the parameters of the virtual model based on the following correction formula:

[0210]

[0211] in, is the modified virtual model, η is the learning rate, which controls the correction amplitude, is the error function The gradient of the virtual model parameters represents the impact of the error on the model parameters.

[0212] The state synchronization generation unit is used to generate key state data consistent with the actual lock state and use it for remote synchronization. This unit integrates the adaptively corrected virtual model with the actual state data to generate key state data for remote monitoring and analysis. Suppose the generated key state data is Used for remote synchronization via network transmission. The state synchronization generation process is implemented using the following formula:

[0213]

[0214] in, is the key status data generated synchronously; W is the weight matrix, which controls the fusion method of the virtual model and the actual data.

[0215] This synchronized data will serve as input to remote monitoring and decision-making systems, facilitating timely status analysis and fault prediction.

[0216] Communication and fault tolerance module

[0217] The communication and fault-tolerance module of the present invention includes a calibration parameter generation unit, a data communication unit, and a fault-tolerant reconstruction unit. The specific implementation is as follows: The calibration parameter generation unit uses the multidimensional time series data collected in real time by various sensors in the ship lock operation monitoring system. The calibration parameter generation unit's main task is to generate redundant calibration parameters for the status data at each time point using the Lie group interpolation method. These redundant calibration parameters are used for fault-tolerant reconstruction of the data.

[0218] The data communication unit is responsible for transmitting the ship status data g(t) and the calibration parameters through the specified data communication protocol. For remote transmission.

[0219] The communication protocol embeds these two parts of data to ensure that even in the event of network interruption or data loss, the lost data can still be restored through the redundancy check parameters. Set the transmission data to It includes state data g(t) and redundancy check parameters

[0220]

[0221] The data communication unit ensures that the data can be transmitted stably in the network. If loss or interruption occurs during the transmission process, it will be restored through the fault-tolerant reconstruction unit.

[0222] The key task of the fault-tolerant reconstruction unit is to ensure that if data is lost or interrupted during network transmission, the system can effectively reconstruct the missing data using Lie group interpolation methods. In the event of data loss or network interruption, the system can reconstruct the missing state data using the redundancy check parameter C(t) and the known state data. Suppose that at a certain time t, the ship's state data is lost. The fault-tolerant reconstruction unit calculates the missing data value using the Lie group interpolation formula.

[0223]

[0224] Among them, g1,g2∈SE(3) represent two known Lie group observations at time points t1 and t2 (usually 4×4 rigid body transformation matrices), describing the spatial pose information, t∈(t1,t2) represents the target time point to be interpolated, which is located between the two observations, and g(t) represents the Lie group element estimated at time t, that is, the interpolation result of the target state on SE(3). represents the inverse matrix of g1, representing the transformation from the reference coordinate system at time t1 to the global or target coordinate system, represents the relative motion from g1 to g2, log(·) represents the logarithmic mapping from Lie group to Lie algebra, Returns the vector or corresponding matrix of 6-dimensional motion. exp(·) represents the exponential mapping from Lie algebra to Lie group. Used to recover the rigid body transformation from the velocity field, represents the normalized interpolation weight of the current time t relative to the two observation moments (ranging from 0 to 1), represents the “motion difference” from g1 to g2 in Lie algebra, and exp(·) represents the interpolation action after scaling the above motion difference and mapping it back to SE(3) space.

[0225] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, so they will not be repeated here.

[0226] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A remote monitoring system for ship lock operation status based on the Internet of Things, characterized in that: The following steps are involved: Data acquisition module, used to obtain multi-source perception data; Edge pre-processing module, used to compress, denoise and unify the format of multi-source perception data; The spatiotemporal alignment and fusion module is used to align and register the processed multi-source data based on their timestamps and spatial pose information, thereby constructing a multi-dimensional fusion tensor. The tensor modeling module, based on the fusion tensor, introduces the geometric constraint information obtained by Lie group space registration, and then constructs a tensor optimization model with spatial structure constraints to extract key tensor features; Fault prediction module, which is used to build a fault propagation path model based on key tensor features and generate predicted status information; The digital twin verification module is used to build a virtual twin system based on predicted state information and compare its operating state with the actual collected data in real time. The prediction model is adaptively corrected through an error feedback mechanism to form a dynamic closed-loop verification mechanism, while also generating key state data for remote synchronization. The communication and fault-tolerance module is used to receive key status data and transmit it remotely through a data communication protocol that includes verification parameters generated based on Lie group interpolation. It also interpolates and reconstructs missing information and maintains synchronization in the event of network interruption or data loss.

2. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The data acquisition module includes: Vibration acquisition unit, used to obtain vibration signals of gates or opening and closing equipment; Water level detection unit, used to obtain water level change data upstream and downstream of the lock; Hydraulic signal acquisition unit, used to collect the working pressure and flow parameters of the opening and closing hydraulic system; Electrical parameter acquisition unit, used to detect voltage, current and power data of the control circuit; The image acquisition unit is used to collect image information of the lock operation area in real time.

3. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The edge preprocessing module includes: Format unification unit, used to convert sensor data of different sources and types into a standard time series format; A data compression unit, used for compressing and encoding multi-source sensing data to reduce transmission bandwidth; The noise suppression unit is used to remove signal noise based on statistical feature filtering or wavelet transform.

4. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The spatiotemporal alignment and fusion module includes: The time alignment unit is used to interpolate and align multi-source data according to their timestamps; The spatial registration unit is used to uniformly map various sensor data to a reference coordinate system using the Lie group registration method based on the spatial posture information of each sensor device; The tensor construction unit is used to fuse the aligned multi-source data into a multi-dimensional tensor structure.

5. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The tensor modeling module includes: A geometric constraint extraction unit, used to extract geometric structure information based on Lie group pose relationships; Spatial structure modeling unit, used to embed geometric constraint information into tensor optimization models; The feature extraction unit is used to extract key low-dimensional features that characterize the operating status of the lock through a tensor decomposition algorithm.

6. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The fault prediction module includes: A state propagation modeling unit, used to build a fault propagation path model based on key tensor features; A state deduction unit is used to deduce the evolution trend of potential system failures based on the constructed path model; The prediction output unit is used to output the prediction status information for use by subsequent modules.

7. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The digital twin verification module includes: A virtual modeling unit, used to construct a digital model of a virtual ship lock based on the predicted state information; A state comparison unit is used to synchronously compare the virtual model's operating state with the actual collected data; Model adaptive correction unit, used to update and adaptively adjust the parameters of the prediction model based on the error feedback mechanism; The state synchronization generation unit is used to generate key state data consistent with the actual state for remote transmission.

8. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The communication and fault-tolerant module includes: A check parameter generating unit, configured to generate redundant check parameters for state data based on a Lie group interpolation method; A data communication unit for remote information transmission according to a data communication protocol embedded with verification parameters; The fault-tolerant reconstruction unit is used to reconstruct missing data based on Lie group interpolation in the event of data interruption or loss.

9. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The tensor modeling module uses the tensor optimization model as follows: Let the fused tensor be X∈R I×J×K , by introducing the geometric constraint matrix G∈R I×R , construct the following optimization objective function: Where X∈R I×J×K Represents the original fusion tensor, which contains multi-source perception data of I, J, and K dimensions, usually corresponding to sensor type, time series, space or feature dimension, G∈R I×R Represents the introduced geometric constraint matrix, which is used to impose a priori constraints on the first factor matrix A in the tensor decomposition, reflecting the spatial structure or posture correlation. R represents the rank of the tensor, which is equivalent to the rank number or number of components of the decomposition, and is the number of low-dimensional vectors used to approximate the tensor. r ∈R I 、b r ∈R J 、c r ∈R K are the decomposition vectors of the rth tensor rank component on three modules, which are used to form a rank tensor. ° represents the outer product operation of the vector, that is, is a rank tensor, Represents the reconstructed tensor expression of the CP decomposition of the tensor, A∈R I×R Indicates that all a r The matrix after the vector column combination, that is, A=[a1,a2,...,a R ], is the factor matrix on the first module, Represents the square of the Frobenius norm, that is, the sum of the squares of all elements of a tensor or matrix, which is used to measure the reconstruction error or matrix deviation, λ∈R + is the regularization coefficient, which is used to balance the influence between the main reconstruction error term and the geometric constraint term.

10. The remote monitoring system for ship lock operation status based on the Internet of Things according to claim 1 is characterized in that: The fault-tolerant reconstruction unit in the communication and fault-tolerant module estimates the missing data using a Lie group interpolation method. The interpolation formula is: Assume that the two adjacent observation values ​​of the key state data on the Lie group are g1, g2∈SE(3), and the time corresponding to the missing state is t∈(t1, t2), then the estimated state is: Among them, g1,g2∈SE(3) represent two known Lie group observations at time points t1 and t2, describing the spatial pose information, t∈(t1,t2) represents the target time point to be interpolated, which is located between the two observations, g(t) represents the Lie group element estimated at time t, that is, the interpolation result of the target state on SE(3), g1 -1 represents the inverse matrix of g1, which represents the transformation from the reference coordinate system at time t1 to the global or target coordinate system, g1 -1 g2 represents the relative motion from g1 to g2, log(·) represents the logarithmic mapping from Lie group to Lie algebra, log:SE(3)→se(3), returns the vector or corresponding matrix of 6-dimensional motion, exp(·) represents the exponential mapping from Lie algebra to Lie group, exp:se(3)→SE(3), used to recover the rigid body transformation from the velocity field, Represents the normalized interpolation weight of the current time t relative to the two observation moments, represents the Lie algebra representation of the motion difference from g1 to g2, and exp(·) represents the interpolation action after scaling the motion difference back to the SE(3) space.

Images