A warehouse management system for intelligent inbound and outbound management

Abstract

This invention relates to the field of automated warehousing and logistics control technology, and more particularly to a warehouse management system for intelligent inbound and outbound management. The system includes: a dynamic response acquisition module that collects vibration data and establishes discrete dynamic state equations; a manifold state mapping module that maps orthogonal mode shapes to Grassmann manifold base points; a tangent space perturbation pre-simulation module that constructs Lie algebraic perturbation operators in the tangent space based on cargo mass and uses a Lie group exponential mapping algorithm to generate predicted manifold states for each candidate storage location; a topological entropy evaluation module that generates a topological health score based on geodesic distance and structural energy entropy indices; a multi-objective optimization decision-making module that selects the optimal storage location based on the score and controls the actuators; and a model parameter self-correction module that calibrates parameters using the deviation between measured and predicted manifold states. This invention achieves real-time pre-simulation and proactive avoidance of the dynamic evolution of the warehouse structure.

Description

Technical Field

[0001] This invention relates to the field of automated warehousing and logistics control technology, and in particular to a warehouse management system for intelligent inbound and outbound management. Background Technology

[0002] With the development of modern logistics towards intensification, ultra-high-rise automated storage and retrieval systems (AS / RS) are widely adopted due to their superior space utilization. These warehouses are typically assembled from thousands of standard steel units, reaching heights of tens of meters, and exhibit significant flexibility and multi-degree-of-freedom vibration characteristics. Existing warehouse management systems primarily allocate storage locations based on cargo turnover rate, operational efficiency, or the static load-bearing capacity of the foundation when scheduling inbound and outbound operations.

[0003] However, traditional systems mostly only perform static load-bearing verification, which ignores the dynamic drift caused by time-varying mass, resulting in early fatigue fracture of critical nodes due to latent resonance. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides an intelligent warehouse management system for inbound and outbound operations, aiming to improve the problem that most traditional systems only perform static load-bearing verification, neglecting the dynamic drift caused by time-varying mass, which leads to early fatigue fracture of key nodes due to latent resonance.

[0005] This invention provides the following technical solution: an intelligent warehouse management system for inbound and outbound management, comprising the following modules:

[0006] The dynamic response acquisition module is used to collect real-time vibration response data of key nodes of the storage structure and establish discrete dynamic state equations.

[0007] The manifold state mapping module is used to extract the orthogonal mode shapes of the state equations and map the subspace spanned by the orthogonal mode shapes to the current manifold base points on the Grassmann manifold.

[0008] The tangent space perturbation pre-simulation module is used to respond to the cargo entry request, calculate the global mass matrix increment based on the cargo mass, construct the corresponding Lie algebra perturbation operator in the tangent space of the current manifold base point, and use the Lie group exponential mapping algorithm to generate the predicted manifold state for each candidate cargo location.

[0009] The topological entropy evaluation module is used to calculate, based on the predicted manifold state, the geodesic distance index, which characterizes the degree of deviation of structural symmetry, and the structural energy entropy index, which characterizes the degree of uniformity of energy distribution, and generate a topological health score accordingly.

[0010] The multi-objective optimization decision-making module is used to select the optimal storage location based on the topology health score and control the action of the actuator;

[0011] The model parameter self-correction module is used to calibrate the system model parameters by utilizing the deviation between the measured manifold state and the predicted manifold state after the operation.

[0012] By adopting the above technical solution, the structural modal evolution is pre-evaluated using the Grassmann manifold tangent space perturbation algorithm, and then modal localization is avoided based on energy entropy. This improves the problem that traditional systems mostly only perform static load verification, which ignores the dynamic drift caused by time-varying mass, resulting in early fatigue fracture of key nodes due to latent resonance.

[0013] Preferably, the dynamic response acquisition module includes:

[0014] The system continuously receives time-series signals uploaded by the accelerometer array and uses the preprocessing logic of the covariance-driven random subspace identification algorithm to construct a Hankel matrix containing the system's historical input and output data.

[0015] Read the current cargo storage distribution data in the warehouse, superimpose the mass of the cargo in each location onto the diagonal of the pre-stored empty mass matrix of the rack structure, and generate a real-time global mass matrix.

[0016] By combining the preset structural damping ratio parameters and the global stiffness matrix, a set of second-order linear differential equations describing the instantaneous motion of the system is assembled, which serves as the discrete dynamic state equation.

[0017] Preferably, the manifold state mapping module includes:

[0018] Singular value decomposition is performed on the Hankel matrix to extract the system observability matrix and state sequence, and then the eigenvalues ​​and eigenvectors of the system state matrix are solved.

[0019] The first few principal mode eigenvectors with energy contribution exceeding a preset threshold are selected, and a modal basis matrix composed of orthogonal column vectors is generated through the Gram-Schmidt orthogonalization process.

[0020] The modal basis matrix is ​​regarded as a plane in a high-dimensional Euclidean space. Using the quotient space projection mapping rule, the plane is mapped to the unique equivalence class coordinate point on the Grassman manifold, which is the current manifold basis point.

[0021] Preferably, the tangent space perturbation pre-simulation module includes:

[0022] Based on the quality of the goods to be put into storage and the node numbers of the candidate storage locations in the finite element model, a local mass increment matrix is ​​constructed that has non-zero values ​​only at the corresponding nodes.

[0023] The modal mass perturbation matrix is ​​obtained by continuously multiplying the transpose of the modal basis matrix corresponding to the current manifold base point with the local mass increment matrix and the original modal basis matrix.

[0024] A weighted coefficient matrix is ​​constructed based on the difference in the frequencies of each modality. The elements in the modal mass perturbation matrix are multiplied by the corresponding weighted coefficients to generate an antisymmetric matrix, which is the Lie algebraic perturbation operator describing the rotational velocity of the tangent space.

[0025] Preferably, the tangent space perturbation pre-simulation module further includes:

[0026] Using the definition of exponential mapping in Lie group theory, the antisymmetric matrix in the tangent space is transformed into a rotation matrix on the orthogonal group;

[0027] The approximate rotation operator is calculated by using the Taylor series expansion method to retain the zeroth-order, first-order, and second-order terms of the rotation matrix.

[0028] The approximate rotation operator is right-multiplied by the modal basis matrix of the current manifold base point to obtain a new basis matrix after rotation transformation. The subspace spanned by this new basis matrix is ​​the predicted manifold state.

[0029] Preferably, the topological entropy evaluation module includes:

[0030] Obtain the reference modal subspace of the structure under ideal periodic symmetry as a reference point;

[0031] Calculate the cross-correlation matrix between the basis matrix of the predicted manifold state and the basis matrix of the reference point, and perform singular value decomposition on the cross-correlation matrix;

[0032] Extract the singular values ​​obtained from the decomposition. These singular values ​​represent the cosine values ​​of the principal angles between the two subspaces. Then, calculate the values ​​of all principal angles using the inverse cosine function.

[0033] Calculate the arithmetic square root of the sum of the squares of all principal angle values, and use it as the geodesic distance index to quantify the length of the geodesic path between two points on the manifold.

[0034] Preferably, the topological entropy evaluation module further includes:

[0035] By calling the element stiffness matrix in the finite element model and combining it with the mode shape vector in the predicted manifold state, the elastic potential energy of each storage unit under each principal mode is calculated.

[0036] The elastic potential energy of each principal mode is weighted and summed according to the reciprocal of the corresponding eigenvalue to obtain the comprehensive modal strain energy of the storage unit.

[0037] Calculate the sum of the comprehensive modal strain energy of all storage units in the entire yard, and divide the comprehensive modal strain energy of a single storage unit by the sum to obtain the energy distribution probability value of that unit;

[0038] Logarithmic operations are performed on the energy distribution probability values ​​of all units, and the weighted sum is taken, with the negative value being used as the structural energy entropy index.

[0039] Preferably, the topological entropy evaluation module further includes:

[0040] The preset historical maximum geodesic distance and maximum theoretical entropy value are obtained respectively, and the calculated geodesic distance index and structural energy entropy index are subjected to dimensionless normalization.

[0041] Establish a linear scoring function that includes a positive gain coefficient and a negative penalty coefficient;

[0042] By substituting the normalized structural energy entropy index into the positive gain term and the normalized geodesic distance index into the negative penalty term, a unique topological health score is calculated.

[0043] Preferably, the multi-objective optimization decision-making module includes:

[0044] Candidate storage locations are sorted in descending order based on their topology health score values ​​to generate an initial selection queue;

[0045] The mechanical stability of the storage locations in the initial queue is checked sequentially. The maximum overturning moment caused by the storage of goods is calculated to be less than the allowable threshold of the structure. The acceleration required for the stacker crane to reach the storage location is also checked to be less than the allowable limit of the equipment.

[0046] The storage location that passes the above verification and has the highest ranking is locked as the optimal storage location. The spatial three-dimensional coordinates of this storage location are then analyzed to generate a motion control message to drive the stacker crane's servo motor.

[0047] Preferably, the model parameter self-correction module includes:

[0048] Construct a likelihood function with the objective of minimizing the geodesic distance between the actual manifold base point and the predicted manifold state;

[0049] By introducing the prior probability distribution of stiffness parameters, a posterior probability density function is constructed.

[0050] The Markov chain Monte Carlo sampling method is used to search the posterior probability density function to obtain the maximum posterior estimate of the stiffness matrix correction coefficient;

[0051] The element stiffness matrix elements in the finite element model inside the system are updated using the maximum a posteriori estimate.

[0052] The present invention has the following beneficial effects:

[0053] 1. In this invention, by using the Grassmann manifold tangent space perturbation algorithm to pre-evaluate the structural modal evolution, and then avoiding modal localization based on energy entropy, the problem of early fatigue fracture of key nodes due to implicit resonance is improved by the fact that most traditional systems only perform static load verification and ignore the dynamic drift caused by time-varying mass due to neglecting the dynamic drift.

[0054] 2. In this invention, by constructing Lie algebraic perturbation operators in the tangent space and using the Lie group exponential mapping algorithm to generate the predicted manifold state, the structural modal evolution can be quickly predicted without re-meshing the mesh. This improves the problem that traditional systems mostly use full-model finite element re-analysis, which is too time-consuming and thus cannot meet the real-time scheduling and response requirements of logistics operations.

[0055] 3. In this invention, by mapping the subspace spanned by orthogonal mode shapes to the base points of the Grassmann manifold and predicting its evolution, the nonlinear frequency drift caused by changes in mass distribution can be accurately perceived. This improves the problem that traditional scheduling mostly uses static bearing capacity verification, which ignores the time-varying dynamic characteristics and thus cannot prevent the structure's natural frequency from drifting to the equipment's excitation range and causing resonance damage.

[0056] 4. In this invention, the deviation between the measured manifold state and the predicted manifold state after operation is used to back-calibrate the system model parameters, thereby eliminating the physical model error caused by structural aging or loose connection. This improves the problem that traditional digital twin models mostly use fixed initial parameters and lack a closed-loop feedback mechanism, resulting in a gradual decrease in prediction accuracy over time. Attached Figure Description

[0057] Figure 1 This is a module architecture diagram of a warehouse management system for intelligent inbound and outbound management proposed in this invention;

[0058] Figure 2 This is a detailed flowchart of the data acquisition and manifold mapping of an intelligent warehouse management system for inbound and outbound management proposed in this invention;

[0059] Figure 3 This invention presents a detailed flowchart of the prediction and decision-making process of a warehouse management system for intelligent inbound and outbound management.

[0060] Figure 4 This is a flowchart illustrating the self-correcting model of a warehouse management system for intelligent inbound and outbound management proposed in this invention. Detailed Implementation

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1:

[0063] In a first embodiment of the present invention, the present invention provides a warehouse management system for intelligent inbound and outbound management, such as... Figures 1-4 As shown, the following systems are included:

[0064] The dynamic response acquisition module is used to collect real-time vibration response data of key nodes of the storage structure and establish discrete dynamic state equations.

[0065] Furthermore, the dynamic response acquisition module includes:

[0066] The system continuously receives time-series signals uploaded by the accelerometer array and uses the preprocessing logic of the covariance-driven random subspace identification algorithm to construct a Hankel matrix containing the system's historical input and output data.

[0067] Read the current cargo storage distribution data in the warehouse, superimpose the mass of the cargo in each location onto the diagonal of the pre-stored empty mass matrix of the rack structure, and generate a real-time global mass matrix.

[0068] By combining the preset structural damping ratio parameters and the global stiffness matrix, a set of second-order linear differential equations describing the instantaneous motion of the system is assembled, which serves as the discrete dynamic state equation.

[0069] Specifically, the dynamic response acquisition module is responsible for the digital conversion of physical signals and the real-time updating of structural parameters, and finally outputs discrete dynamic state equations.

[0070] The module receives multi-channel vibration signals from an array of accelerometers at the connection points between the rack uprights and beams. Since environmental excitations are typically assumed to be white noise, the module employs preprocessing logic based on a covariance-driven random subspace identification algorithm to process these time-series signals. The input is... One sensor in Output vector of sampling at time step The module arranges the output data according to time delay and constructs a block Hankel matrix. This matrix not only contains information about the current vibration amplitude but also implies the system's correlation characteristics over time. The Hankel matrix involved... The construction formula is as follows:

[0071] ;

[0072] in Represents the covariance sequence of the output signal. For the number of rows in the block, The number of blocks. Specifically defined as ,in For mathematical expectation operators, For a moment The output data vector, This indicates the transpose operation. By constructing this matrix, the module transforms the originally discrete and disordered time series into an algebraic form containing system modal information, providing standardized data input for the subsequent manifold state mapping module to extract orthogonal mode shapes.

[0073] Unlike traditional fixed-parameter models, this module needs to respond to the flow of stored goods. The module reads the current goods storage distribution data from the warehouse management database, which includes the coordinates of the storage locations and the corresponding mass values ​​of the goods. The module internally stores the initial mass matrix of the racking structure in an unloaded state. This matrix is ​​established based on the physical properties of the finite element mesh nodes. When goods are stored or removed, the module will determine the quality of the goods. Mapping the data to the corresponding finite element node of the cargo location, the diagonal elements of the initial mass matrix are corrected to generate the time-varying real-time global mass matrix. The formula for correcting the mass matrix is:

[0074] ;

[0075] in for The real-time global quality matrix at time t. The structure's unloaded mass matrix. The total number of occupied storage spaces. For warehouse location index, For the first The quality of goods in each storage location To map the local storage location node degrees of freedom to the global structural degrees of freedom, a Boolean positioning matrix is ​​used. This indicates a diagonal matrix transformation operation. This step ensures that the subsequently established dynamic equations accurately reflect the structural inertial characteristics under the current load.

[0076] Obtain the real-time global quality matrix Then, the module calls the preset structural damping ratio parameters to construct the damping matrix. It also calls the global stiffness matrix determined by the structural material properties and geometry. The module assembles the three matrix parameters mentioned above to establish a system of second-order linear differential equations describing the instantaneous motion of the system under environmental excitation. The discrete dynamic state equations are expressed as follows:

[0077] ;

[0078] in Let be the displacement vector of the structural node. The velocity vector of the structural nodes. Let be the acceleration vector of the structural node. This represents the external environmental excitation force vector. The output terminal passes the coefficient matrix of this equation system to the manifold state mapping module. This parametric modeling approach ensures that subsequent manifold analysis is based on a real physical model, rather than relying solely on the statistical regularities of pure data. This guarantees that the model maintains high fidelity even when changes in cargo cause the system's inherent frequency to drift.

[0079] The manifold state mapping module is used to extract the orthogonal mode shapes of the state equations and map the subspace spanned by the orthogonal mode shapes to the current manifold base points on the Grassmann manifold.

[0080] Furthermore, the manifold state mapping module includes:

[0081] Singular value decomposition is performed on the Hankel matrix to extract the system observability matrix and state sequence, and then the eigenvalues ​​and eigenvectors of the system state matrix are solved.

[0082] The first few principal mode eigenvectors with energy contribution exceeding a preset threshold are selected, and a modal basis matrix composed of orthogonal column vectors is generated through the Gram-Schmidt orthogonalization process.

[0083] The modal basis matrix is ​​regarded as a plane in a high-dimensional Euclidean space. Using the quotient space projection mapping rule, the plane is mapped to the unique equivalence class coordinate point on the Grassman manifold, which is the current manifold basis point.

[0084] Specifically, the manifold state mapping module, as a computational unit connecting the physical vibration space and the geometric manifold space, receives the block Hankel matrix output by the dynamic response acquisition module at its input end and generates the coordinates of the Grassmann manifold base point, representing the current dynamic characteristics of the system, at its output end. This module, through feature extraction and spatial mapping, compresses the complex modal information of a high-dimensional time-varying system into a geometric point on a low-dimensional manifold space.

[0085] Module receive block Hankel matrix Singular value decomposition is then performed on the signal to separate the signal and noise subspaces. The decomposition process follows the mathematical expression below:

[0086]

[0087] in It is a left singular vector matrix. It is a diagonal matrix containing singular values. It is a right singular vector matrix. Represents transpose. Based on the descending order of singular values, the module truncates the result before... The submatrices corresponding to the larger singular values ​​are used to construct the system's observability matrix. By utilizing the least squares method or pseudo-inverse operation, combined with the block structure of time-shift characteristics, the system state matrix of the discrete state-space model can be solved. Therefore, the eigenvalue problem equation is established: ;in For the eigenvalues ​​of a discrete system, These are the corresponding eigenvectors. A logarithmic transformation is then performed on the eigenvalues. The module calculates the natural frequency and damping ratio of the physical system, where This represents the sampling time interval.

[0088] The module calculates the energy contribution of each mode and filters out modes whose energy contribution is greater than a preset threshold. The former The principal modal eigenvectors of the first order form the original modal matrix. To satisfy the definition of a Grassmann manifold, the basis vectors spanning its subspace must be mutually orthogonal. (Module pair) The column vectors are subjected to Gram-Schmidt orthogonalization. For the ... Original feature vectors The orthogonalized vector The calculation is as follows:

[0089]

[0090] in Representing vectors In vector Projection on For vector norm, These are the final generated unit orthogonal basis vectors. After processing, the module output is... The modal basis matrix consists of orthogonal column vectors. ,satisfy ,in It is an identity matrix.

[0091] The module will use orthogonal modal basis matrices Considered Void space one of the A linear subspace. According to the definition of a Grassmann manifold, all orthogonal basis matrices spanning the same subspace belong to the same equivalence class. Using the quotient space projection rule, the module will... Mapped to Grassmann manifold The unique equivalence class coordinate point on The mapping relationship is expressed as follows:

[0092] ;

[0093] in for The orthogonal group of order 1 represents a subspace whose rotational transformations of basis vectors do not change the positions of points on the manifold. Through this step, the module eliminates the interference of coordinate selection on the system state description, outputting the current manifold base points. It becomes the geometric reference origin for subsequent tangent space perturbation pre-simulation modules.

[0094] The tangent space perturbation pre-simulation module is used to respond to the cargo entry request, calculate the global mass matrix increment based on the cargo mass, construct the corresponding Lie algebra perturbation operator in the tangent space of the current manifold base point, and use the Lie group exponential mapping algorithm to generate the predicted manifold state for each candidate cargo location.

[0095] Furthermore, the tangent space perturbation pre-simulation module includes:

[0096] Based on the quality of the goods to be put into storage and the node numbers of the candidate storage locations in the finite element model, a local mass increment matrix is ​​constructed that has non-zero values ​​only at the corresponding nodes.

[0097] The modal mass perturbation matrix is ​​obtained by continuously multiplying the transpose of the modal basis matrix corresponding to the current manifold base point with the local mass increment matrix and the original modal basis matrix;

[0098] A weighted coefficient matrix is ​​constructed based on the difference in the frequencies of each modality. The elements in the modal mass perturbation matrix are multiplied by the corresponding weighted coefficients to generate an antisymmetric matrix. This antisymmetric matrix is ​​the Lie algebraic perturbation operator describing the rotational velocity of the tangent space.

[0099] The tangent space perturbation pre-simulation module also includes:

[0100] Using the definition of exponential mapping in Lie group theory, the antisymmetric matrix in the tangent space is transformed into a rotation matrix on the orthogonal group;

[0101] The approximate rotation operator is obtained by using the Taylor series expansion method to preserve the zeroth, first, and second-order terms of the rotation matrix.

[0102] Multiply the approximate rotation operator on the right by the modal basis matrix of the current manifold base point to obtain a new basis matrix after rotation transformation. The subspace spanned by this new basis matrix is ​​the predicted manifold state.

[0103] Specifically, the tangent space perturbation pre-simulation module, as the prediction core of the system, is responsible for quickly simulating the impact of cargo storage on the structural dynamics without re-performing the global finite element analysis. This module receives the mass parameters of the cargo to be stored, the coordinates of candidate storage locations, and the current manifold base points output by the manifold state mapping module at its input end. Its output end generates the predicted manifold state for each candidate storage location.

[0104] The module first reads the weight value of the goods to be received into the warehouse. Determine the node number corresponding to the candidate storage location in the finite element model. Based on this, the module constructs a local mass increment matrix. This matrix is ​​sparse, existing only in the nodes. The relevant degrees of freedom have non-zero values ​​in the column and row. All other elements are zero. Then, the module calls the modal basis matrix corresponding to the current manifold base point. Perform continuous matrix multiplication to calculate the modal mass perturbation matrix. The calculation formula is: ;in It is the transpose of the modal basis matrix. elements in Characterized the first First-order mode and second-order mode The coupling strength between first-order modes caused by the increase in mass. Obtain the matrix. Then, based on the first-order perturbation theory, the module combines the eigenvalues ​​corresponding to each mode. Construct an antisymmetric matrix This refers to the Lie algebra perturbation operator. Antisymmetric matrix. The Middle Line 1 Column elements The calculation formula is: ;in and The first Rank and number eigenvalues ​​of order 1 This represents the element at the corresponding position in the modal mass perturbation matrix. This operator... Lie algebra space belonging to the orthogonal group Geometrically, this describes the rotational tendency and rate of the structural state on the tangent plane of the current manifold base point.

[0105] The module utilizes Lie group theory to map linear transformations in the tangent space back to the curved manifold space. According to the definition of exponential mapping, the rotation matrix on the orthogonal group... With Lie algebra elements The relationship is To avoid the high computational cost of matrix exponentiation operations, the module uses a second-order Taylor series expansion method to calculate the approximate rotation operator. The calculation formula is:

[0106] in It is the identity matrix. The antisymmetric matrix generated in the preceding steps is used. This formula preserves the zeroth, first, and second-order components of the rotation transformation, significantly reducing computational complexity while maintaining accuracy. Finally, the module approximates the rotation operator. Right multiply by the current modal basis matrix The updated basis matrix is ​​obtained. The updated formula is: ; by the new basis matrix The linear subspace spanned by the column vectors of the candidate storage location is defined as the predictive manifold state for that candidate storage location. This process enables a rapid forward derivation from physical mass parameters to geometric manifold coordinates, providing a mathematical object containing structural evolution information for subsequent topology evaluation.

[0107] The topological entropy assessment module is used to calculate the geodesic distance index, which characterizes the degree of deviation of structural symmetry, and the structural energy entropy index, which characterizes the degree of uniformity of energy distribution, based on the predicted manifold state, and generate a topological health score accordingly.

[0108] Furthermore, the topological entropy evaluation module includes:

[0109] Obtain the reference modal subspace of the structure under ideal periodic symmetry as a reference point;

[0110] Calculate the cross-correlation matrix between the basis matrix of the predicted manifold state and the basis matrix of the reference point, and perform singular value decomposition on the cross-correlation matrix;

[0111] Extract the singular values ​​obtained from the decomposition. These singular values ​​represent the cosine values ​​of the principal angles between the two subspaces. Then, calculate the values ​​of all principal angles using the inverse cosine function.

[0112] Calculate the arithmetic square root of the sum of the squares of all principal angle values, and use it as the geodesic distance index to quantify the length of the geodesic path between two points on the manifold.

[0113] The topological entropy evaluation module also includes:

[0114] The element stiffness matrix in the finite element model is called, and the modal shape vector in the predicted manifold state is combined to calculate the elastic potential energy of each storage unit under each principal mode.

[0115] The elastic potential energy of each principal mode is weighted and summed according to the reciprocal of the corresponding eigenvalue to obtain the comprehensive modal strain energy of the storage unit.

[0116] Calculate the sum of the comprehensive modal strain energy of all storage units in the entire yard, and divide the comprehensive modal strain energy of a single storage unit by the sum to obtain the energy distribution probability value of that unit;

[0117] Logarithmic operations are performed on the energy distribution probability values ​​of all units, and the weighted sum is taken, with the negative value being used as the structural energy entropy index.

[0118] The topological entropy evaluation module further includes:

[0119] The preset historical maximum geodesic distance and maximum theoretical entropy value are obtained respectively, and the calculated geodesic distance index and structural energy entropy index are subjected to dimensionless normalization.

[0120] Establish a linear scoring function that includes a positive gain coefficient and a negative penalty coefficient;

[0121] By substituting the normalized structural energy entropy index into the positive gain term and the normalized geodesic distance index into the negative penalty term, a unique topological health score is calculated.

[0122] Specifically, the topology entropy evaluation module, as the system's quantitative evaluation unit, receives the predicted manifold state generated by the tangent space perturbation pre-simulation module and pre-stored ideal reference state data at its input end, and generates a unique topology health score to guide scheduling at its output end. This module performs dual verification of the safety of candidate cargo locations from two dimensions: macroscopic symmetry and microscopic energy distribution, by calculating geometric distance and physical energy entropy.

[0123] The module first calls the reference mode subspace in the ideal periodic symmetric state from memory, and denotes its corresponding orthogonal basis matrix as follows: Simultaneously, obtain the basis matrix of the predicted manifold state corresponding to the current candidate storage location. The module calculates the cross-correlation matrix between two basis matrices. The calculation formula is Then the matrix Perform singular value decomposition. The decomposition formula is expressed as:

[0124] ;

[0125] in and It is an orthogonal matrix. For containing singular values The diagonal matrix. Extract the diagonal matrix. The singular value in the equation is geometrically equal to the principal angle between two higher-dimensional subspaces. The cosine value. The module uses the inverse cosine function. The module calculates a set of principal angle values. Based on the Riemannian geometric definition, it calculates the arithmetic square root of the sum of the squares of all principal angle values ​​to obtain the geodesic distance index. The calculation formula is as follows: ;in Let be the dimension of the subspace. This metric quantifies the degree to which the current predicted state deviates from the ideal periodic structure.

[0126] The module calls the element stiffness matrix from the finite element model database. For each storage unit The module utilizes the first manifold state prediction in the prediction manifold state. First-order mode shape vector The elastic potential energy of the unit in this mode is calculated. Subsequently, the module introduces an eigenvalue weighting mechanism. The first... Eigenvalues ​​corresponding to the first mode Calculate the combined modal strain energy of this element. The calculation formula is as follows:

[0127] ;

[0128] in This is the transpose of the mode shape vector. The reciprocal of the eigenvalues ​​is used as a weighting coefficient to enhance the influence of low-frequency modes on the energy distribution. This yields all... After calculating the comprehensive modal strain energy of each storage unit, the module calculates the total energy and... And calculate the energy distribution probability value of a single cell. The probability is defined as: ;

[0129] Based on information entropy theory, the module analyzes probability sequences. Perform calculations and output the structural energy entropy index. The calculation formula is: ;in It is the natural logarithm. This index reflects the degree of dispersion of vibration energy distribution in the structural space.

[0130] The module reads the preset historical maximum geodesic distance in the system. With the maximum theoretical entropy value For the calculated and Dimensionless normalization is performed. The module establishes a linear scoring function, including a positive gain coefficient. With negative penalty coefficient The scoring formula is as follows:

[0131] ;

[0132] in This is the final output topology health score. Positive terms reward solutions with uniform energy distribution, while negative terms penalize solutions that break structural symmetry. This score is directly passed to the multi-objective optimization decision-making module as the ranking basis for cargo location selection.

[0133] The multi-objective optimization decision-making module is used to select the optimal storage location based on the topology health score and control the actions of the actuators.

[0134] Furthermore, the multi-objective optimization decision-making module includes:

[0135] Candidate storage locations are sorted in descending order based on their topology health score values ​​to generate an initial selection queue;

[0136] The mechanical stability of the storage locations in the initial queue is checked sequentially. The maximum overturning moment caused by the storage of goods is calculated to be less than the allowable threshold of the structure. The acceleration required for the stacker crane to reach the storage location is also checked to be less than the allowable limit of the equipment.

[0137] The storage location that passes the above verification and has the highest ranking is locked as the optimal storage location. The spatial three-dimensional coordinates of this storage location are then analyzed to generate a motion control message to drive the stacker crane's servo motor.

[0138] Specifically, the multi-objective optimization decision-making module, acting as the system's logical control terminal, connects the computational layer and the physical execution layer. Its input receives a dataset containing the topological health scores of all candidate storage locations, generated by the topological entropy evaluation module. Its output sends underlying motion control messages to the stacker crane servo drive system. This module ensures that the final decision strikes a balance between theoretical optimality and engineering feasibility through sorting, filtering, and multiphysics constraint verification.

[0139] The module reads the topological health score sequence from memory. The processor executes a quicksort algorithm, sorting all candidate locations in descending order based on their score values. After sorting, a list of candidate location indices is formed in memory, with the location at the top of the list representing the best overall performance in terms of manifold geometric distance and energy entropy. This step establishes a priority order based on inherent structural safety, transforming complex manifold analysis results into a linear decision-making criterion.

[0140] The module reads the storage locations in the initial queue sequentially and performs rigid body static verification. For the currently verified storage location... The module reads its height coordinates. and the quality of goods to be stored The system calculates the overturning moment of the entire warehouse relative to the base fulcrum after the goods are stored. The calculation formula is as follows:

[0141] ;

[0142] in The cumulative overturning moment under the current inventory status. This is the gravitational acceleration constant. The module calls the preset allowable structural torque threshold. Execute the judgment logic:

[0143] ;

[0144] If the inequality does not hold, it indicates that although the storage location has a high topology score, it may lead to macroscopic structural instability. The module immediately removes the storage location and reads the next location in the queue for verification. This step prevents the risk of pursuing modal equilibrium while neglecting fundamental mechanical safety.

[0145] After passing the mechanical stability verification, the module proceeds to the equipment capability verification stage. The module calculates how the stacker crane can move from its current position to the target storage location. Required kinematic parameters. Focus on verifying jerk. This refers to the rate of change of acceleration, used to assess the impact on the equipment. Assume the planned acceleration time is... The target acceleration is The acceleration is estimated as follows:

[0146] ;

[0147] The module reads the rated jerk limit of the stacker crane servo system. Execute the judgment logic:

[0148] ;

[0149] If the calculated value exceeds the limit, it indicates that the scheduling task requires the equipment to operate too violently, which may cause motor overload or cargo slippage. The system will also remove the storage location.

[0150] When a storage location in the queue simultaneously satisfies the above-mentioned mechanical stability and equipment motion calibration tests, the module locks it as the final optimal storage location. The processor then analyzes the three-dimensional spatial coordinates of this storage location in the warehouse coordinate system. According to the stacker crane control protocol, the module encapsulates coordinate data and action commands into hexadecimal control messages. These messages are sent to the stacker crane's programmable logic controller (PLC) via a fieldbus or Ethernet interface. After parsing the messages, the PLC drives the horizontal travel motor, lifting motor, and fork extension motor to perform precise closed-loop position control, completing the physical storage and retrieval actions.

[0151] The model parameter self-correction module is used to calibrate the system model parameters by utilizing the deviation between the measured manifold state and the predicted manifold state after the operation.

[0152] Furthermore, the model parameter self-correction module includes:

[0153] Construct a likelihood function with the objective of minimizing the geodesic distance between the actual manifold base point and the predicted manifold state;

[0154] By introducing the prior probability distribution of stiffness parameters, a posterior probability density function is constructed.

[0155] The Markov chain Monte Carlo sampling method is used to search the posterior probability density function to obtain the maximum posterior estimate of the stiffness matrix correction coefficient;

[0156] The element stiffness matrix elements in the internal finite element model of the system are updated using the maximum a posteriori estimate.

[0157] Specifically, the model parameter self-correction module, as the closed-loop feedback calibration unit of the system, is executed after the physical access operation is completed. This module receives the actual manifold baseline points generated by the manifold state mapping module based on the current steady-state vibration acquisition, and the predicted manifold states historically generated by the tangent space perturbation pre-simulation module at its input. At its output, it generates corrected stiffness matrix correction coefficients, which are used to update the finite element model parameter library within the system.

[0158] The module first quantifies the difference between model predictions and physical measurements. It then retrieves the actual manifold baselines after the task is completed. The corresponding predicted manifold state generated before the operation ,in This represents the stiffness parameter vector to be corrected. Based on the geometric properties of the Grassmann manifold, the module calculates the geodesic distance between them. Based on the assumption that the measurement error follows a Gaussian distribution, a likelihood function is constructed. The calculation formula is as follows:

[0159] ;

[0160] in For geodesic distance function, This measures the standard deviation of the noise. A larger value for this function indicates a higher current stiffness parameter. This makes the model's predictions closer to the actual physical state.

[0161] The module incorporates a Bayesian inference framework. Stiffness parameters are defined based on the engineering tolerance range of the shelving material properties. Prior probability distribution This is typically assumed to be a normal distribution with a mean of 1. Combining the likelihood function and the prior distribution, the module constructs the posterior probability density function. The relationships are expressed as follows:

[0162] ;

[0163] The denominator is a normalization constant. This function integrates prior engineering knowledge with current measured data, characterizing the probability distribution of stiffness parameters under known observation data conditions.

[0164] Because the posterior probability density function involves complex nonlinear manifold mappings, it cannot be solved analytically directly. The module employs the Metropolis-Hastings algorithm from the Markov chain Monte Carlo sampling method to search the parameter space. Starting from the initial state, the module randomly generates candidate parameters. And based on the probability of acceptance Decide whether to update the status. The formula for calculating the acceptance probability is:

[0165] ;

[0166] After a sufficient number of iterative sampling iterations, the chain converges to a stationary distribution. The module statistically analyzes the sampling results and extracts the parameter value corresponding to the peak value of the posterior probability density function, i.e., the maximum posterior estimate. .

[0167] Module parsing maximum a posteriori estimate Each element in the vector The corresponding structure Stiffness correction coefficients for element types. The module calls the finite element model database to adjust the element stiffness matrix. Perform an update operation. The update formula is as follows:

[0168] ;

[0169] in The initial stiffness matrix, This is the corrected stiffness matrix. After the update, the new global stiffness matrix will be used in the next round of dynamic response acquisition and tangent space pre-simulation, realizing the synchronous mapping of stiffness degradation caused by physical system aging or loose connections in the digital model.

[0170] Example 2:

[0171] In the application scenarios of ultra-high-rise heavy-duty automated storage and retrieval systems (AS / RS), the racking structure is typically tens of meters high and exhibits significant flexibility. With the high-frequency random access of heavy industrial components, the global mass distribution matrix of the structure exhibits drastic time-varying characteristics. This dynamic evolution of mass distribution not only causes nonlinear drift in the structure's natural frequencies, easily coupling with the excitation frequencies generated by the high-speed operation and start-stop of stacker cranes, leading to resonance, but also induces modal localization due to the uneven distribution of goods disrupting the periodic symmetry of the structure. This results in vibration energy being exponentially confined to a few local elements, causing hidden early fatigue failure. However, most existing warehouse scheduling systems rely solely on static load-bearing capacity verification, lacking the ability to perceive the aforementioned dynamic evolution risks. Furthermore, traditional full-model finite element reanalysis methods are computationally time-consuming and cannot meet the millisecond-level real-time scheduling response requirements of logistics operations, thus posing a serious structural safety hazard. To solve the above problems, this invention provides an intelligent warehouse management system for inbound and outbound management, the structure of which is as follows: Figure 1 As shown. The specific implementation process of this system is as follows:

[0172] The dynamic response acquisition module, in conjunction with the manifold state mapping module, achieves the characteristic transformation from physical vibration signals to geometric manifold space. The dynamic response acquisition module establishes discrete dynamic state equations, transforming the instantaneous motion of the storage structure under complex environmental excitations into a computable mathematical model, providing a physical benchmark for analysis. The manifold state mapping module abandons the traditional one-sided analysis based on a single frequency or mode shape, mapping the subspace spanned by orthogonal mode shapes to base points on the Grassmann manifold. This approach eliminates the interference of coordinate system selection on modal analysis, compressing the complex structural dynamic state into a single object in a unified geometric space, thus providing mathematical rigor and geometric intuition for subsequent measurement and calculation of the structural evolution trajectory.

[0173] The tangent space perturbation pre-simulation module overcomes the problem of excessive computation time and inability to meet the millisecond-level scheduling requirements of traditional finite element reanalysis methods. This module utilizes Lie group and Lie algebra theory to transform the nonlinear structural property changes caused by goods entering the warehouse into linear algebraic operations on the tangent plane of the manifold. By constructing a Lie algebra perturbation operator and combining it with an exponential mapping algorithm, the system can quickly deduce the evolution position of the structural modal subspace after changes in mass distribution without re-meshing or reassembling the global stiffness matrix. This method significantly reduces computational complexity while maintaining prediction accuracy, ensuring that the scheduling system can perform real-time pre-simulations of thousands of candidate schemes.

[0174] The topological entropy assessment module addresses the technical limitation of modal localization phenomena, which are difficult to monitor using conventional methods. By calculating the geodesic distance index, the system can quantify the degree to which the current structural state deviates from the ideal periodic symmetry state and identify macroscopic dynamic characteristic drift. By calculating the structural energy entropy index, the system introduces information theory methods to analyze the full-field strain energy distribution and accurately identify local elements with highly concentrated vibration energy. The comprehensively generated topological health score transforms abstract structural safety into comparable numerical criteria, enabling the scheduling system to proactively avoid loading positions that, while meeting static bearing capacity requirements, may induce early fatigue fracture.

[0175] The multi-objective optimization decision-making module realizes the transformation from passive adaptation to active control. Selecting the optimal storage location based on topological health scores essentially treats the inventory as a tuning mass block, utilizing its physical properties to actively improve the dynamic balance of the structure and suppress modal localization trends. The model parameter self-correction module constitutes the system's closed-loop feedback mechanism. Using post-operation measured data, the system's internal model parameters are calibrated in reverse, eliminating deviations between the theoretical model and the actual structure caused by structural aging, loose connections, or construction errors, ensuring the system's predictive accuracy and robustness during long-term operation.

[0176] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A warehouse management system for intelligent inbound and outbound management, characterized in that, Includes the following modules: The dynamic response acquisition module is used to collect real-time vibration response data of key nodes of the storage structure and establish discrete dynamic state equations. The manifold state mapping module is used to extract the orthogonal mode shapes of the state equations and map the subspace spanned by the orthogonal mode shapes to the current manifold base points on the Grassmann manifold. The tangent space perturbation pre-simulation module is used to respond to the cargo entry request, calculate the global mass matrix increment based on the cargo mass, construct the corresponding Lie algebra perturbation operator in the tangent space of the current manifold base point, and use the Lie group exponential mapping algorithm to generate the predicted manifold state for each candidate cargo location. The topological entropy evaluation module is used to calculate, based on the predicted manifold state, the geodesic distance index, which characterizes the degree of deviation of structural symmetry, and the structural energy entropy index, which characterizes the degree of uniformity of energy distribution, and generate a topological health score accordingly. The multi-objective optimization decision-making module is used to select the optimal storage location based on the topology health score and control the action of the actuator; The model parameter self-correction module is used to calibrate the system model parameters by utilizing the deviation between the measured manifold state and the predicted manifold state after the operation.

2. The intelligent warehouse management system for inbound and outbound management according to claim 1, characterized in that, The dynamic response acquisition module includes: The system continuously receives time-series signals uploaded by the accelerometer array and uses the preprocessing logic of the covariance-driven random subspace identification algorithm to construct a Hankel matrix containing the system's historical input and output data. Read the current cargo storage distribution data in the warehouse, superimpose the mass of the cargo in each location onto the diagonal of the pre-stored empty mass matrix of the rack structure, and generate a real-time global mass matrix. By combining the preset structural damping ratio parameters and the global stiffness matrix, a set of second-order linear differential equations describing the instantaneous motion of the system is assembled, which serves as the discrete dynamic state equation.

3. The intelligent warehouse management system for inbound and outbound management according to claim 1, characterized in that, The manifold state mapping module includes: Singular value decomposition is performed on the Hankel matrix to extract the system observability matrix and state sequence, and then the eigenvalues ​​and eigenvectors of the system state matrix are solved. The first few principal mode eigenvectors with energy contribution exceeding a preset threshold are selected, and a modal basis matrix composed of orthogonal column vectors is generated through the Gram-Schmidt orthogonalization process. The modal basis matrix is ​​regarded as a plane in a high-dimensional Euclidean space. Using the quotient space projection mapping rule, the plane is mapped to the unique equivalence class coordinate point on the Grassman manifold, which is the current manifold basis point.

4. The intelligent warehouse management system for inbound and outbound management according to claim 1, characterized in that, The tangent space perturbation pre-simulation module includes: Based on the quality of the goods to be put into storage and the node numbers of the candidate storage locations in the finite element model, a local mass increment matrix is ​​constructed that has non-zero values ​​only at the corresponding nodes. The modal mass perturbation matrix is ​​obtained by continuously multiplying the transpose of the modal basis matrix corresponding to the current manifold base point with the local mass increment matrix and the original modal basis matrix. A weighted coefficient matrix is ​​constructed based on the difference in the frequencies of each modality. The elements in the modal mass perturbation matrix are multiplied by the corresponding weighted coefficients to generate an antisymmetric matrix, which is the Lie algebraic perturbation operator describing the rotational velocity of the tangent space.

5. A warehouse management system for intelligent inbound and outbound management according to claim 1, characterized in that, The tangent space perturbation pre-simulation module also includes: Using the definition of exponential mapping in Lie group theory, the antisymmetric matrix in the tangent space is transformed into a rotation matrix on the orthogonal group; The approximate rotation operator is calculated by using the Taylor series expansion method to retain the zeroth-order, first-order, and second-order terms of the rotation matrix. The approximate rotation operator is right-multiplied by the modal basis matrix of the current manifold base point to obtain a new basis matrix after rotation transformation. The subspace spanned by this new basis matrix is ​​the predicted manifold state.

6. The intelligent warehouse management system for inbound and outbound management according to claim 1, characterized in that, The topology entropy evaluation module includes: Obtain the reference modal subspace of the structure under ideal periodic symmetry as a reference point; Calculate the cross-correlation matrix between the basis matrix of the predicted manifold state and the basis matrix of the reference point, and perform singular value decomposition on the cross-correlation matrix; Extract the singular values ​​obtained from the decomposition. These singular values ​​represent the cosine values ​​of the principal angles between the two subspaces. Then, calculate the values ​​of all principal angles using the inverse cosine function. Calculate the arithmetic square root of the sum of the squares of all principal angle values, and use it as the geodesic distance index to quantify the length of the geodesic path between two points on the manifold.

7. A warehouse management system for intelligent inbound and outbound management according to claim 1, characterized in that, The topological entropy evaluation module also includes: By calling the element stiffness matrix in the finite element model and combining it with the mode shape vector in the predicted manifold state, the elastic potential energy of each storage unit under each principal mode is calculated. The elastic potential energy of each principal mode is weighted and summed according to the reciprocal of the corresponding eigenvalue to obtain the comprehensive modal strain energy of the storage unit. Calculate the sum of the comprehensive modal strain energy of all storage units in the entire yard, and divide the comprehensive modal strain energy of a single storage unit by the sum to obtain the energy distribution probability value of that unit; Logarithmic operations are performed on the energy distribution probability values ​​of all units, and the weighted sum is taken. The negative value is used as the structural energy entropy index.

8. A warehouse management system for intelligent inbound and outbound management according to claim 1, characterized in that, The topological entropy evaluation module further includes: The preset historical maximum geodesic distance and maximum theoretical entropy value are obtained respectively, and the calculated geodesic distance index and structural energy entropy index are subjected to dimensionless normalization. Establish a linear scoring function that includes a positive gain coefficient and a negative penalty coefficient; By substituting the normalized structural energy entropy index into the positive gain term and the normalized geodesic distance index into the negative penalty term, a unique topological health score is calculated.

9. A warehouse management system for intelligent inbound and outbound management according to claim 1, characterized in that, The multi-objective optimization decision-making module includes: Candidate storage locations are sorted in descending order based on their topology health score values ​​to generate an initial selection queue; The mechanical stability of the storage locations in the initial queue is checked sequentially. The maximum overturning moment caused by the storage of goods is calculated to be less than the allowable threshold of the structure. The acceleration required for the stacker crane to reach the storage location is also checked to be less than the allowable limit of the equipment. The storage location that passes the above verification and has the highest ranking is locked as the optimal storage location. The spatial three-dimensional coordinates of this storage location are then analyzed to generate a motion control message to drive the stacker crane's servo motor.

10. A warehouse management system for intelligent inbound and outbound management according to claim 1, characterized in that, The model parameter self-correction module includes: Construct a likelihood function with the objective of minimizing the geodesic distance between the actual manifold base point and the predicted manifold state; By introducing the prior probability distribution of stiffness parameters, a posterior probability density function is constructed. The Markov chain Monte Carlo sampling method is used to search the posterior probability density function to obtain the maximum posterior estimate of the stiffness matrix correction coefficient; The element stiffness matrix elements in the finite element model inside the system are updated using the maximum a posteriori estimate.

Images