Method for assessing the load-bearing capacity of the roof slab of passenger and freight elevators with adaptive early warning function

Abstract

This invention provides a method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function. The method includes the following steps: collecting data from multiple sensor sources and employing a dynamic synchronous sampling optimization algorithm for data acquisition and preprocessing; extracting time-frequency features and using a multi-dimensional time-frequency analysis algorithm to separate load modes and construct a dynamic load spectrum; establishing a multi-scale state mapping and using a generalized dynamic response analysis method to calculate the system response and assess structural risk; constructing an adaptive early warning system, dynamically optimizing the early warning threshold, and generating the optimal decision-making scheme. This method can adaptively adjust the sampling strategy, accurately identify load characteristics, achieve multi-scale state assessment, and dynamically optimize early warning decisions, significantly improving the accuracy and reliability of the load-bearing capacity assessment of the roof slab of passenger / freight elevators.

Description

Technical Field

[0001] This invention relates to load-bearing capacity assessment, and in particular to a method for assessing the load-bearing capacity of the roof slab of a passenger and freight elevator with adaptive early warning function. Background Technology

[0002] With the rapid development of urban construction, the demand for passenger and freight elevators in building construction is increasing daily. As a critical structure supporting construction personnel and materials, the load-bearing capacity of the passenger and freight elevator roof directly affects construction safety and efficiency. Especially during the construction of super high-rise buildings, the passenger and freight elevator roof needs to withstand complex and variable load conditions, including the combined effects of dynamic impact loads, asymmetric load distribution, and temperature stress. Therefore, developing a roof load-bearing capacity assessment method with adaptive early warning capabilities is of significant practical importance for preventing safety accidents, ensuring the safety of construction personnel and property, and improving construction efficiency.

[0003] Currently, the load-bearing capacity assessment of passenger and freight elevator roof slabs mainly employs traditional static analysis methods, such as linear finite element analysis and empirical formula calculations. Some researchers have begun to introduce dynamic response analysis, using accelerometers to monitor the vibration characteristics of the roof slab and assessing the structural state through spectral analysis. Other studies use strain gauges to obtain stress distribution data and combine this with modal analysis to identify structural damage. In terms of early warning, single-indicator monitoring based on fixed thresholds is mainly used, such as displacement over-limit alarms and stress over-limit alarms. Some researchers have attempted to use artificial intelligence methods such as neural networks to build early warning models, but these are primarily based on offline-trained static models.

[0004] However, existing technical solutions still suffer from the following problems: First, sensor data acquisition lacks an adaptive optimization mechanism, and the sampling frequency is fixed, making it difficult to dynamically adjust according to changes in operating conditions, potentially leading to the loss of important data features at critical moments. Second, the load feature extraction process does not fully consider time-varying characteristics, especially under non-stationary operating conditions, where traditional time-frequency analysis methods struggle to accurately capture the dynamic changes in load patterns. Third, most state assessment models are based on single-scale analysis, neglecting the coupling effect between macroscopic structural response, local stress distribution, and material property evolution, making it difficult to comprehensively reflect the structural state. Fourth, early warning decision-making lacks adaptive capabilities; early warning thresholds are set too rigidly and fail to dynamically adjust based on real-time risk assessment results. Furthermore, the decision optimization process does not fully consider multi-objective trade-offs, making it difficult to achieve an optimal balance between efficiency, timeliness, and cost. Finally, system parameter optimization lacks an effective feedback learning mechanism, failing to automatically adjust and optimize system performance based on historical decision-making results. These technical problems severely restrict the practical application of the load-bearing capacity assessment system for passenger and freight elevator roof slabs. Summary of the Invention

[0005] The purpose of this invention is to provide a method for assessing the load-bearing capacity of the roof slab of a passenger and freight elevator with adaptive early warning function, in order to solve the aforementioned problems existing in the prior art.

[0006] The technical solution, a method for assessing the load-bearing capacity of the roof slab of a passenger and freight elevator with adaptive early warning function, includes the following steps:

[0007] Step S1: Collect data streams from multiple sensors, including load sensor data, vibration sensor data, displacement sensor data, and temperature sensor data, and preprocess them to obtain the final processed data stream;

[0008] Step S2: Receive the final processed data stream, construct a time-frequency window function based on the data time series, convolve the window function with the data stream, and calculate the time-frequency energy distribution matrix based on the extracted time-frequency features; separate different load modes from the time-frequency energy distribution matrix, calculate the correlation coefficient between each mode, and construct a mode weight matrix; evaluate the energy contribution rate of each load mode, calculate its coefficient of variation and stability index, generate a feature importance score, perform feature filtering and fusion based on the score, and obtain a fused feature vector; calculate the load probability density distribution based on the fused feature vector, construct a dynamic load spectrum, and establish a reliability index.

[0009] Step S3: Receive the dynamic load spectrum and reliability indicators, and input these data into the macroscopic structural response analysis module, local stress distribution analysis module, and material performance analysis module respectively to calculate the state vectors at three levels. Establish a state transition matrix based on the distance relationship between the three levels of state vectors. Input the state transition matrix into the dynamic equation solver to calculate the acceleration response spectrum, velocity response spectrum, and displacement response spectrum of the system. Calculate the static risk value, dynamic risk value, and fatigue risk value based on the response spectrum data, and determine the weighting coefficients according to the information entropy of each risk indicator to generate a comprehensive risk indicator and risk evolution trend. Predict the future state distribution based on the risk evolution trend, calculate the probability of the system entering a failure state, and obtain the reliability prediction value and failure probability prediction value.

[0010] Step S4: Receive reliability prediction values, failure probability prediction values, and comprehensive risk indicators; calculate the initial warning threshold based on historical data; construct a dynamic threshold function by combining the warning threshold with the risk change rate; calculate the warning discrimination value based on the dynamic threshold function and dynamic response characteristics; obtain the warning level by combining the warning discrimination value with the prior probability distribution; construct a decision space based on the warning level and risk evolution trend; calculate the benefit value, timeliness index, and cost index of each optional decision scheme; obtain the optimal decision scheme through multi-objective optimization; compare the execution effect of the optimal decision scheme with the expected goal, calculate the evaluation error, and update the knowledge base parameters and system parameters based on the error.

[0011] The beneficial effects include the ability to adaptively adjust sampling strategies, accurately identify load characteristics, achieve multi-scale state assessment, dynamically optimize early warning decisions, and significantly improve the accuracy and reliability of load-bearing capacity assessment for passenger and freight elevator roofs. Attached Figure Description

[0012] Figure 1 This is a flowchart of the present invention.

[0013] Figure 2 This is a flowchart of step S1 of the present invention.

[0014] Figure 3 This is a flowchart of step S2 of the present invention.

[0015] Figure 4 This is a flowchart of step S3 of the present invention.

[0016] Figure 5 This is a flowchart of step S4 of the present invention. Detailed Implementation

[0017] Example 1, such as Figure 1 As shown, the method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function includes the following steps:

[0018] Step S1: Collect data streams from multiple sensors, including load sensor data, vibration sensor data, displacement sensor data, and temperature sensor data. Calculate the importance score for each sensor based on its signal strength, noise level, and signal change rate. Adjust the sampling frequency of each sensor based on its importance score to obtain a synchronized and optimized data stream. Decompose the synchronized and optimized data stream into multi-dimensional harmonic components and residual terms. Calculate the local variance of each frequency component, construct a filtering function, and filter the data to obtain a denoised data stream. Map the denoised data stream to a multi-dimensional manifold space, calculate the gradient vector in the manifold space, and identify abnormal data based on the local average eigenvector and eigenvector standard deviation to obtain a data stream with anomaly markers. Construct a local reconstruction function based on the anomaly-marked data stream, calculate the reconstruction error, update the compensation coefficients, and obtain the final processed data stream.

[0019] Step S2: Receive the final processed data stream output from Step S1; construct a time-frequency window function based on the data time series; convolve the window function with the data stream; calculate the time-frequency energy distribution matrix based on the extracted time-frequency features; separate different load modes from the time-frequency energy distribution matrix; calculate the correlation coefficient between each mode; construct a mode weight matrix; evaluate the energy contribution rate of each load mode; calculate its coefficient of variation and stability index; generate a feature importance score; perform feature filtering and fusion based on the score to obtain a fused feature vector; calculate the load probability density distribution based on the fused feature vector; construct a dynamic load spectrum; and establish a reliability index.

[0020] Step S3: Receive the dynamic load spectrum and reliability index output from step S2, and input these data into the macroscopic structural response analysis module, local stress distribution analysis module, and material performance analysis module respectively to calculate the state vectors at three levels; establish a state transition matrix based on the distance relationship between the three levels of state vectors; input the state transition matrix into the dynamic equation solver to calculate the acceleration response spectrum, velocity response spectrum, and displacement response spectrum of the system; calculate the static risk value, dynamic risk value, and fatigue risk value based on the response spectrum data, and determine the weighting coefficients according to the information entropy of each risk index to generate a comprehensive risk index and risk evolution trend; predict the future state distribution based on the risk evolution trend and calculate the probability of the system entering a failure state.

[0021] Step S4: Receive the reliability prediction value, failure probability prediction value, and comprehensive risk index output from step S3; calculate the initial warning threshold based on historical data; construct a dynamic threshold function by combining the warning threshold with the risk change rate; calculate the warning discrimination value based on the dynamic threshold function and dynamic response characteristics; obtain the warning level by combining the warning discrimination value with the prior probability distribution; construct a decision space based on the warning level and risk evolution trend; calculate the benefit value, timeliness index, and cost index of each optional decision scheme; obtain the optimal decision scheme through multi-objective optimization; compare the execution effect of the optimal decision scheme with the expected goal; calculate the evaluation error; and update the knowledge base parameters and system parameters based on the error.

[0022] This technical solution achieves intelligent and adaptive assessment of elevator roof load-bearing capacity by constructing a complete technical chain of "data acquisition - feature extraction - state assessment - early warning decision-making". The innovation of the solution is mainly reflected in: ① adaptive acquisition and intelligent preprocessing of multi-source data, ensuring the quality of raw data; ② the application of time-frequency joint analysis and mode decomposition technology, improving the accuracy of feature extraction; ③ the construction of a multi-scale state assessment model, realizing comprehensive state monitoring; ④ the design of an adaptive early warning decision-making mechanism, ensuring the timeliness of early warning and the optimality of decision-making. These technological innovations work together to form a closed-loop intelligent assessment system, significantly improving the accuracy, reliability, and practicality of elevator roof load-bearing capacity assessment. In practical applications, this solution can effectively prevent safety accidents, protect the lives and property of construction personnel, and improve construction efficiency, demonstrating significant engineering application value.

[0023] According to one aspect of this application, step S1 specifically comprises:

[0024] Step S11: According to the preset reference sampling frequency, acquire load sensor data, vibration sensor data, displacement sensor data and temperature sensor data respectively, calculate the signal strength, noise level and signal change rate of each sensor, multiply these parameters by preset weight coefficients and sum them to obtain the sensor importance score, calculate the optimal sampling frequency based on the sensor importance score, and resample the original data stream using the optimal sampling frequency to obtain the synchronously optimized data stream.

[0025] The adaptive sampling frequency optimization method is defined as follows: f(i,t) = f_base·exp(η(t)·W(i,t))·(1 + δ(t)·ΔW(i,t)); where f(i,t) is the optimal sampling frequency of sensor i at time t; f_base is the reference sampling frequency; and η(t) = η o ·(1-exp(-λ·||dW / dt||)) is the adaptive gain coefficient; W(i,t) is the sensor importance score; ΔW(i,t) = W(i,t) - W̄(t) is the relative importance deviation, W̄(t) is the average importance score; δ(t) = δ o ·exp(-μ·var(W(:,t))) is the compensation coefficient; η o , λ, δ o μ are control parameters; ||·|| is the Euclidean norm; var(·) is the variance operator.

[0026] Step S12: Receive the synchronized and optimized data stream output in step S11, divide the data stream into multiple time windows according to the time series, apply the generalized harmonic basis function to the data in each time window to decompose a series of harmonic components and corresponding residual terms; calculate the variance value of each harmonic component within its local time window, construct an adaptive filtering function based on the variance value; apply the adaptive filtering function to each harmonic component, perform threshold filtering on the residual terms, and recombine the processed harmonic components and residual terms to obtain the denoised data stream.

[0027] Step S13: Import the denoised data stream obtained in step S12, establish a multi-dimensional coordinate system, and map the data points into this coordinate system to form a manifold space; calculate the Euclidean distance between adjacent data points and construct a local connectivity matrix; calculate the local gradient vector of each point based on the connectivity matrix; calculate the average eigenvector and standard deviation of the eigenvector within the local region based on historical data; compare the current gradient vector with the average eigenvector and determine outliers based on the preset standard deviation threshold; add marking information to the identified outliers and output the data stream with outlier markings.

[0028] Step S14: Read the data stream with anomaly markers output in step S13, and divide the data sequence into multiple overlapping local intervals according to time order; within each local interval, select a suitable combination of basis functions based on the distribution characteristics of non-anomalies; calculate the weight coefficients of the basis functions using the least squares method to establish a local reconstruction function; calculate the mean square error between the reconstruction function and the actual data; iteratively update the weight coefficients based on the gradient information of the mean square error until the preset convergence condition is met; correct and compensate for the data of anomalies using the reconstruction function; smoothly connect the reconstruction results of all local intervals, and finally output the processed data stream.

[0029] By constructing a sensor importance scoring function and an adaptive sampling frequency optimization mechanism, intelligent acquisition and synchronous optimization of multi-source sensor data are achieved. This step employs wavelet transform for data denoising, combines manifold learning for anomaly detection, and finally corrects and compensates for anomalous data through a local reconstruction function. This multi-level data preprocessing method significantly improves the quality and reliability of the original data. Specifically, in the scenario of elevator roof load-bearing capacity assessment, it can effectively handle noise interference and anomalous data from load sensors, vibration sensors, displacement sensors, and temperature sensors, ensuring the accuracy of subsequent analysis. By dynamically adjusting the sampling frequency, the system can increase the sampling density at critical moments (such as sudden load changes) and decrease the sampling frequency in a stable state, ensuring data integrity while optimizing the efficiency of system resource utilization.

[0030] According to one aspect of this application, step S2 specifically comprises:

[0031] Step S21: Import the final processed data stream output from step S1, segment the data stream, and apply a preset time window function to weight each segment; construct an adaptive kernel function, which includes a time-domain window function and a phase compensation function; perform convolution operation between the weighted data segments and the adaptive kernel function to calculate the time-frequency distribution matrix; extract the amplitude information and phase information from the time-frequency distribution matrix respectively; calculate the time-frequency energy distribution based on the amplitude information and phase information to form a time-frequency feature matrix.

[0032] The process of time-frequency feature extraction is as follows: TF(t,ω) = |∫x(τ)·g((t-τ) / α(t))·exp(-jωτ+ jφ(t,ω))dτ| 2 Where: TF(t,ω) is the time-frequency energy distribution; x(τ) is the signal time series; g(·) is the adaptive window function, g(u) = exp(-u 2 / 2)·(1 + β(t)·u 4 ); α(t) is the time-varying window width parameter; φ(t,ω) = θ(t)·ω 2 + γ(t)·ω³ is the phase compensation function; β(t), θ(t), and γ(t) are dynamic optimization parameters; t is the time variable; ω is the frequency variable; and j is the imaginary unit.

[0033] Step S22: Read the time-frequency feature matrix output in step S21, construct the time-varying amplitude function and frequency phase function, and combine these functions to form the load mode basis functions; calculate the projection coefficients of the time-frequency feature matrix on each basis function; sort the load modes according to the magnitude of the projection coefficients and retain the main load modes; calculate the inner product between the retained load modes to obtain the modal correlation coefficient matrix; determine the weight coefficients of each mode according to the correlation coefficient matrix and the modal energy values; combine the weight coefficients with the corresponding load modes to output the load mode set.

[0034] The load mode decomposition process is as follows: B_k(t,ω) = A_k(t)·exp(jΦ_k(ω))·exp(-ζ_k(t)·|ω-ω_k(t)| 2 ); where: B_k(t,ω) is the time-frequency basis function of the k-th load mode; A_k(t) is the time-varying amplitude function, obtained through EMD decomposition; Φ_k(ω) is the frequency phase function, calculated using the phase unwinding algorithm; ζ_k(t) is the frequency localization parameter; ω_k(t) is the center frequency function; k is the mode number; t is the time variable; ω is the frequency variable; |·| is the absolute value operator.

[0035] Step S23: Receive the load mode set output in step S22, calculate the energy value, variance, and time series stability index of each mode; multiply the energy value, variance, and stability index by preset weight coefficients to obtain the importance score of each mode; calculate the redundancy penalty factor based on the correlation coefficient between modes; combine the importance score with the redundancy penalty factor to determine the final feature selection criterion; screen modes based on the selection criterion, and weight the retained modes according to the importance score to generate a fused feature vector.

[0036] Step S24: Import the fused feature vector output in step S23, and segment the feature vector within a preset time window; calculate statistical features for each segment of data, including the mean vector and covariance matrix; construct a multidimensional Gaussian mixture model based on the mean vector and covariance matrix, and calculate the probability density distribution of the load; differentiate the probability density distribution function to obtain the load change rate; construct a dynamic load spectrum based on the load change rate; calculate the system reliability index based on the dynamic load spectrum, and output the dynamic load spectrum and reliability index.

[0037] By introducing time-frequency joint analysis and mode decomposition techniques, multi-dimensional extraction and fusion of load-bearing characteristics of elevator roof slabs were achieved. This step first uses time-frequency transformation to obtain the time-frequency feature matrix of the signal, then uses mode decomposition algorithms to separate different load modes, and performs feature selection and fusion based on energy contribution rate and stability indices. This method can not only capture the static characteristics of the load but also reflect its dynamic changes. In practical applications, the system can accurately identify the load characteristics of elevator roof slabs under different working conditions, such as static load, dynamic impact, and fatigue load, and by constructing a dynamic load spectrum, it provides a reliable feature basis for subsequent state assessment. Especially when dealing with non-stationary working conditions, this method can effectively capture the dynamic changes of load modes, improving the accuracy and adaptability of feature extraction.

[0038] According to one aspect of this application, step S3 specifically comprises:

[0039] Step S31: Import the dynamic load spectrum and reliability index output in step S2, calculate the overall structural response vector using the finite element model, calculate the local stress distribution vector using the stress concentration factor, and generate the microscopic performance vector by combining the material property curves; calculate the Euclidean distance between adjacent state vectors and construct a state transition function with a time decay term; dynamically adjust the bandwidth parameter of the state transition function according to the rate of change of the reliability index; apply the adjusted state transition function to each state vector to generate a state transition probability matrix; apply a time weight to the state transition probability matrix and output the state mapping matrix.

[0040] The multi-scale state mapping is specifically: Q(t) = exp(-||S(t)-S_ref||2 / θ(t))·exp(λ(t)·R(t))·M(t); where: Q(t) is the state mapping matrix; S(t) is the current state vector; S_ref is the reference state vector; θ(t) = θ o ·exp(-η·|dR / dt|) is the adaptive bandwidth function; R(t) is the reliability index; λ(t) is the state weight function; M(t) is the quality matrix; ||·|| is the Mahalanobis distance; θ o η is the control parameter; t is the time variable.

[0041] Step S32: Read the state mapping matrix output in step S31, construct a time-varying mass matrix based on the system mass distribution; construct a damping matrix based on the nonlinear relationship between displacement and velocity; calculate the stiffness matrix by combining material properties and geometric features; substitute these matrices into the dynamic equations; construct a set of orthogonal dynamic basis functions, and calculate the combination coefficients of the basis functions using the variational principle; solve the dynamic equations through iterative optimization to obtain time-domain response data; perform Fourier transform on the time-domain response data to obtain the acceleration response spectrum, velocity response spectrum, and displacement response spectrum, respectively.

[0042] The solution method for the dynamic equation is: x** + C(x,x*)·x* + K(x)·x = F(t); where: C(x,x*) = C o + α·|x*| + β·x 2 The nonlinear damping matrix; K(x) = K o + γ·x + δ·x³ is the nonlinear stiffness matrix; F(t) is the external load vector; C o K is the linear damping coefficient; o α is the linear stiffness coefficient; α, β, γ, and δ are nonlinear parameters; x is the displacement vector; x* is the velocity vector; x** is the acceleration vector; and t is the time variable.

[0043] Step S33: Receive the three types of response spectrum data output in step S32; calculate the static over-limit probability based on the extreme value distribution theory to obtain the static risk index; calculate the dynamic impact effect using the time integral of the response spectrum to obtain the dynamic risk index; calculate the cumulative fatigue damage value based on the cycle number and amplitude distribution of the response spectrum to obtain the fatigue risk index; calculate the time series entropy values ​​of the three risk indices; determine the weighting coefficient of each risk index based on the entropy value; substitute the weighted risk indices into the risk evolution equation; solve the risk evolution equation using the numerical integration method to obtain the comprehensive risk index and risk evolution trend.

[0044] The risk assessment process is as follows: R(t) = w1(t)·R_s(t) + w2(t)·R_d(t) + w3(t)·R_f(t); where: R(t) is the comprehensive risk index; R_s(t) = P(max|x(t)| > x_cr) is the static over-limit risk, calculated based on the extreme value distribution theory; R_d(t) = ∫|x**(t)|dt / T_cr is the dynamic impact risk; R_f(t) = Σ(n_i / N_i)^m is the fatigue damage risk; w_k(t) = exp(H_k(t)) / Σexp(H_j(t)) is the entropy weight coefficient; H_k(t) is the information entropy; x_cr is the critical displacement; T_cr is the characteristic time; n_i is the number of cycles; N_i is the lifetime number of cycles; and m is the material constant.

[0045] Step S34: Import the comprehensive risk index and risk evolution trend output in step S33, and establish a prediction model describing the system state transition; construct a risk impact function based on the risk index; introduce a random disturbance term into the state prediction model to simulate the random fluctuation characteristics of the system; define the safety domain boundary conditions of the system; generate a large number of state samples using the Monte Carlo method; statistically analyze the proportion of samples located within the safety domain to obtain the reliability prediction value; calculate the probability of the sample first crossing the safety domain boundary to obtain the failure probability prediction value; output the reliability prediction value and the failure probability prediction value.

[0046] The reliability prediction process is as follows: P_f(t,τ) = ∫∫p(z,ω)·I(g(z,ω) < 0)dzdω·exp(-λ(t)·R(t)); where: P_f(t,τ) is the predicted failure probability; p(z,ω) is the state-load joint probability density; g(z,ω) is the limit state function; I(·) is the characteristic function; z is the state vector; ω is the load vector; λ(t) is the time-varying decay coefficient; R(t) is the reliability index; t is the current time; and τ is the prediction time step.

[0047] By constructing a multi-scale state assessment model and using a nonlinear dynamic response analysis method, a comprehensive assessment and risk prediction of the load-bearing state of the elevator roof slab was achieved. This step innovatively introduces coupled analysis at three levels: macroscopic structural response, local stress distribution, and material properties, and describes the interactions between these different levels using a state transition matrix. Simultaneously, a dynamic equation solver, combined with the entropy weight method, is used for risk assessment, enabling accurate prediction of the system's reliability and failure probability. This multi-scale analysis method is particularly suitable for critical structures like elevator roof slabs that bear complex loads, as it can simultaneously consider multiple aspects such as overall deformation, local stress concentration, and material property degradation, providing more comprehensive and accurate state assessment results.

[0048] According to one aspect of this application, step S4 specifically comprises:

[0049] Step S41: Import the reliability prediction value, failure probability prediction value, and comprehensive risk index output in step S3; read the initial threshold for each level from the early warning level database; calculate the time derivative of the comprehensive risk index to obtain the risk change rate; substitute the risk change rate into the exponential decay function to calculate the threshold adjustment factor; multiply the initial threshold by the adjustment factor to generate the dynamic early warning threshold; calculate the mean square error between the current early warning threshold and the historical best threshold to obtain the threshold credibility; construct constraint equations based on the threshold credibility and obtain the target threshold through numerical solution; use the gradient descent method to iteratively optimize the early warning threshold until the convergence condition is met, and output the dynamic early warning threshold set and threshold credibility.

[0050] The dynamic threshold optimization process is as follows: Th(k,t) = Th o (k)·exp(-γ(t)·|dR / dt|)·(1+ η(t)·ΔR(t)); where: Th(k,t) is the k-level warning threshold; Th o (k) is the initial threshold; γ(t) = γ o ·(1-exp(-α·||dR / dt||)) is the adjustment factor; η(t) is the compensation coefficient; ΔR(t) is the risk change; R(t) is the comprehensive risk index; k is the warning level; t is the time variable; γ o α is a control parameter.

[0051] Step S42: Read the dynamic warning threshold set output in step S41 and the dynamic response features output in step S3; obtain the historical statistical distribution of each response feature from the feature database; calculate the deviation between the current response feature and the warning threshold; calculate the alarm weight coefficient of each feature based on the deviation; substitute the weighted feature deviation into the discriminant function to obtain the warning discriminant value; read the prior probability distribution from the warning level database; calculate the probability density of the warning discriminant value under each level condition; calculate the posterior probability distribution according to Bayes' theorem; select the level with the highest posterior probability as the current warning level; calculate the entropy value of the posterior probability distribution to obtain the level stability index; and output the warning level and level stability.

[0052] The specific process for determining the warning level is as follows: L(t) = argmax_k{π_k·exp(-||W(t)-W_k|| 2 / σ 2 _k(t))};where: L(t) is the warning level; π_k is the prior probability; W(t) is the current warning feature vector; W_k is the standard feature vector of level k; σ 2_k(t) is the dynamic variance parameter; k is the rank number; t is the time variable; ||·|| is the Mahalanobis distance.

[0053] Step S43: Receive the warning level and level stability output from step S42, and the risk evolution trend output from step S3; read the optional alarm strategies, maintenance suggestions, and response measures from the decision database; construct a decision benefit evaluation function, and substitute the warning level, level stability, and risk evolution trend into the function to calculate the benefit value; calculate the timeliness index based on the availability of maintenance resources and scheduling constraints; estimate the cost index based on manpower, material resources, and time consumption; substitute the benefit value, timeliness index, and cost index into the objective function; construct decision constraints, including resource limitations and operating procedure requirements; solve the constrained optimization problem using the Lagrange multiplier method to obtain the optimal decision scheme and optimization index.

[0054] The decision optimization process is as follows: J(d,t) = λ1·E(d,t) + λ2·T(d,t) + λ3·C(d,t); where: J(d,t) is the objective function; E(d,t) = β1·R(t)·S(t) + β2·dR / dt is the benefit function; T(d,t) =exp(-α·t_d) is the time-efficiency function; C(d,t) = c o + c1·||d|| + c2·||d|| 2 denoted as the cost function; d is the decision vector; R(t) is the risk index; S(t) is the level stability; t_d is the decision delay; and λ_i, β_i, α, and c_i are weight coefficients.

[0055] Step S44: Import the optimal decision scheme and optimization index output in step S43; obtain real-time feedback data on decision execution from the monitoring system; calculate the deviation between the actual execution effect and the optimal decision scheme; construct decision evaluation index; substitute the deviation into the evaluation function to calculate the evaluation score; read system parameters and historical evaluation data from the knowledge base; calculate the knowledge update gradient based on the evaluation score; substitute the gradient information into the parameter update equation; obtain new system parameters through iterative calculation; store the updated system parameters and execution effect data into the knowledge base; output the updated knowledge base and optimized system parameters.

[0056] The parameter self-learning process is specifically as follows: θ(t+1) = θ(t) - η(t)·grad J(θ(t))·exp(-μ·E(t)); where: θ(t) is the system parameter vector; η(t) is the adaptive learning rate; J(θ) is the performance objective function; E(t) =||d*(t) - d_real(t)|| 2d*(t) represents the execution error; grad represents the gradient operator; d*(t) represents the optimal decision; d_real(t) represents the actual execution effect; μ represents the decay coefficient; and t represents the time variable.

[0057] By establishing a dynamic early warning threshold and a multi-objective decision optimization mechanism, the adaptiveness and optimal decision-making of elevator roof load-bearing capacity early warning are achieved. This step innovatively proposes a dynamic threshold adjustment method based on the risk change rate and constructs an early warning level assessment model by combining historical data and real-time monitoring results. In the decision optimization stage, by considering factors such as benefits, timeliness, and cost through multi-objective trade-offs, intelligent and precise early warning decisions are achieved. The system also possesses self-learning capabilities, continuously optimizing system parameters and the knowledge base based on the decision execution effect, thereby improving the adaptability and reliability of the early warning system. This adaptive early warning mechanism is particularly suitable for scenarios like elevator roofs, where safety requirements are high and operating conditions are complex, enabling timely detection of potential risks and provision of optimal solutions.

[0058] According to one aspect of this application, step S11 specifically comprises:

[0059] Step S111: According to the preset initial sampling frequency of the sensors, raw data are collected from the load sensor, vibration sensor, displacement sensor and temperature sensor respectively, and the data stream of each sensor is timestamped and calibrated; the data stream is reordered according to the time sequence relationship of the timestamps; the calibration curve of each sensor is read from the sensor parameter library and the raw data is converted into physical quantities; dimensional normalization processing is performed on each data stream, and the normalized multi-source sensor data stream is output.

[0060] Step S112: Import the normalized data stream output in step S111, divide the data into data segments according to a preset time window; calculate the signal variance for each data segment to obtain the signal strength coefficient; calculate the noise level coefficient of the data segment using the multi-scale permutation entropy algorithm; calculate the signal change rate based on the LZ complexity algorithm; multiply these three coefficients by the preset weight coefficients read from the weight database; perform feature space mapping on the weighted coefficients and output the importance score value of each sensor.

[0061] Step S113: Receive the sensor importance score output in step S112, read the benchmark sampling frequency from the sampling parameter library; construct a frequency adjustment equation based on the Sigmoid function; substitute the importance score into the frequency adjustment equation; introduce a dynamic compensation term, which adaptively adjusts according to the score difference between adjacent sensors; obtain the optimal sampling frequency for each sensor through iterative calculation; quantize the sampling frequency to ensure that it meets the hardware sampling capability constraints, and output the optimized sampling frequency for each sensor.

[0062] Step S114: Read the optimized sampling frequency output in step S113 and re-acquire data from each sensor; apply anti-aliasing filtering based on wavelet transform to the newly acquired data; interpolate and align the filtered data according to the minimum time interval; construct a time synchronization evaluation function to calculate the phase difference between data streams; use a dynamic time warping algorithm to eliminate the phase difference; perform linear interpolation on the aligned data to ensure that all data streams have a uniform time step; output the synchronized optimized data stream.

[0063] By constructing a multi-layered sensor data acquisition and optimization mechanism, intelligent and efficient data acquisition is achieved. This step first establishes a sensor importance scoring function based on signal strength, noise level, and signal change rate, dynamically adjusting the sampling frequency to ensure sufficiently dense data acquisition at critical moments. Then, quantization processing and anti-aliasing filtering techniques are employed to guarantee data synchronization and integrity. This step is particularly important in elevator roof load-bearing capacity assessment scenarios because it adaptively adjusts the data acquisition strategy according to the characteristics of different sensors and the importance of the data. For example, when a sudden load change or abnormal vibration is detected, the system automatically increases the sampling frequency of relevant sensors to ensure the capture of key dynamic features. Phase differences are eliminated through a dynamic time warping algorithm, achieving precise alignment of multi-source data and providing a high-quality data foundation for subsequent feature extraction and state assessment. This intelligent data acquisition mechanism significantly improves the system's response speed and data quality while optimizing the efficiency of system resource utilization.

[0064] According to one aspect of this application, step S14 specifically comprises:

[0065] Step S141: Receive the data stream with anomaly markers and read the smoothing window length parameter from the signal analysis parameter library; divide the data stream into multiple overlapping local time intervals based on the rate of change of the data; calculate the Fourier spectrum of the data in each interval; determine the main frequency component of each interval based on the spectral analysis results; select the basis function combination corresponding to the main frequency component from the basis function library; perform orthogonalization on each basis function to form a complete local basis function set; output the local interval division result and the corresponding basis function set.

[0066] Step S142: Import the local interval partitioning results and basis function set output in step S141, and read the error evaluation parameters from the data processing parameter library; calculate the projection coefficients of non-abnormal data points in each local interval onto the basis function set; construct a local reconstruction function based on the projection coefficients; substitute the reconstruction function into the original data to calculate the residuals; evaluate the residuals using the Huber loss function; construct an optimization objective function based on the gradient information of the residuals; iteratively optimize the projection coefficients using the L-BFGS algorithm; and output the optimized reconstruction function parameters.

[0067] Step S143: Read the reconstruction function parameters output in step S142, extract the time location information of outliers from the outlier database; calculate the theoretical value of the outlier location based on the reconstruction function; construct an adaptive weighting function that considers the time distance and amplitude difference between outliers and nearby normal points; apply the weighting function between the theoretical value and the actual value; use the Kalman filtering algorithm to correct the outliers; calculate the local consistency index of the corrected data; output the corrected and compensated data sequence.

[0068] Step S144: Receive the corrected and compensated data sequence output in step S143; calculate the overlapping portion of adjacent local intervals; construct a transition function based on improved Hermite interpolation; apply the transition function to the overlapping intervals for data fusion; calculate the gradient continuity index of the fused region; adaptively adjust the parameters of the transition function according to the continuity index; perform final smoothing processing on the entire data sequence; verify the overall smoothness and local fidelity of the processed data; output the final processed data stream.

[0069] By constructing a data correction and compensation mechanism based on local reconstruction, intelligent processing of abnormal data and improvement of signal quality are achieved. This step innovatively introduces a local interval partitioning method based on Fourier spectral analysis, combined with the Huber loss function and L-BFGS optimization algorithm, to achieve accurate reconstruction of abnormal data. In the scenario of elevator roof load-bearing capacity assessment, this method can effectively handle outliers in various sensor data, such as data anomalies caused by load mutations and sensor failures. By employing the Kalman filtering algorithm for outlier correction and combining it with a transition function of improved Hermite interpolation for data fusion, the continuity and smoothness of the reconstructed data are ensured. This technological innovation not only improves data reliability but also preserves the intrinsic characteristics of the signal, providing high-quality data support for subsequent feature extraction and state assessment. Especially when processing data under highly dynamic operating conditions, this method demonstrates superior robustness and adaptability.

[0070] According to one aspect of this application, step S22 specifically comprises:

[0071] Step S221: Import the time-frequency feature matrix and determine the number of modal basis functions based on the singular value distribution of the matrix; construct a generalized Gabor basis function considering time-varying characteristics; orthogonalize the basis functions using the Schmidt orthogonalization method; calculate the time-frequency localization characteristic index of the basis functions; optimize the basis function parameters based on the characteristic index; obtain the optimal basis function set through iterative calculation; output the optimized modal basis function set.

[0072] Step S222: Receive the modal basis function set output in step S221 and construct a multi-resolution analysis framework based on wavelet packets; calculate the projection coefficients of the signal and basis functions at each resolution level; perform sparse decomposition using the matching pursuit algorithm; perform time-frequency consistency verification on the decomposition results; select the main projection components based on the consistency verification results; sort the selected projection coefficients according to their energy magnitude; and output an ordered set of modal projection coefficients.

[0073] Step S223: Read the modal projection coefficient set output in step S222, construct an instantaneous correlation analysis method based on Hilbert transform; calculate the instantaneous phase difference between each mode; use the mutual information algorithm to evaluate the nonlinear correlation between modes; construct the modal correlation matrix; calculate the eigenvalue decomposition of the correlation matrix; determine the number of independent modes based on the eigenvalue magnitude; output the modal correlation evaluation results.

[0074] Step S224: Import the modal correlation evaluation results output in step S223, and adopt the weight initialization method based on information entropy; construct a weight optimization objective function considering modal independence; apply the alternating direction multiplier method to solve the weight optimization problem; calculate the stability index of the optimized weights; fine-tune the weights according to the stability index; combine the adjusted weights with the corresponding modes; and output the load mode set.

[0075] By constructing a mode decomposition framework based on generalized Gabor basis functions and multi-resolution analysis, accurate extraction of load features and mode identification were achieved. This step first determines the number of modal basis functions based on singular value decomposition and optimizes the characteristics of the basis functions using the Schmidt orthogonalization method. Then, sparse decomposition is performed using the matching pursuit algorithm, combined with instantaneous correlation analysis using Hilbert transform, ultimately achieving accurate separation and feature extraction of load modes. In the scenario of elevator roof load-bearing capacity assessment, this method can effectively identify and separate different types of load modes, such as static loads, dynamic impacts, and periodic vibrations. A weight optimization strategy based on information entropy ensures the reliability and stability of the mode decomposition results. This technological innovation significantly improves the accuracy and resolution of load feature extraction, providing a reliable feature foundation for subsequent state assessment.

[0076] According to one aspect of this application, step S32 specifically comprises:

[0077] Step S321: Receive the state mapping matrix and read the mass distribution data from the system parameter library; construct the time-varying mass matrix using the Newmark integral format; construct the nonlinear damping matrix based on the coupling relationship between velocity and displacement; read the constitutive relation data from the material parameter library; construct the geometric stiffness matrix using the piecewise linearization method; introduce the influence factors of ambient temperature and humidity into the stiffness calculation; correct the singular values ​​of the stiffness matrix; output the mass matrix, damping matrix, and stiffness matrix of the system.

[0078] Step S322: Read the system matrix output in step S321, construct dynamic basis functions based on the Krylov subspace method; determine the main components of the basis functions using SVD decomposition; construct generalized basis functions considering higher-order modes; calculate the orthogonality index of the basis functions; orthogonalize the basis functions using the Gram-Schmidt method; evaluate the computational complexity of the basis functions; optimize the number of basis functions according to complexity and accuracy requirements; output the optimized dynamic basis function set.

[0079] Step S323: Import the dynamic basis function set output in step S322, construct a displacement interpolation function based on the variational principle; discretize the dynamic equations using the Galerkin method; construct an iterative solution format considering nonlinear terms; design an adaptive time step control strategy; solve the nonlinear equation set using the predictive correction method; calculate the residual vector for each time step; dynamically adjust the iteration parameters according to the residual magnitude; output the time domain response data.

[0080] Step S324: Receive the time-domain response data output in step S323, and use the Welch method to estimate the power spectrum; construct a multi-window time-frequency analysis framework; calculate the frequency response function of the system; extract transient response features using wavelet transform; resample the response data in the frequency domain; calculate the acceleration spectrum, velocity spectrum, and displacement spectrum respectively; smooth the response spectrum; and output the three types of response spectra.

[0081] By constructing a framework for solving nonlinear dynamic equations, high-precision analysis of the dynamic response of elevator roof slabs was achieved. This step innovatively introduced a hybrid algorithm based on Newmark integrals and Krylov subspaces, combined with the Galerkin method for equation discretization. By designing an adaptive time-step control strategy and a predictive correction method, the solution accuracy and computational efficiency were significantly improved. In the scenario of elevator roof slab load-bearing capacity assessment, this method can accurately capture the dynamic response characteristics of the structure under complex loads, including acceleration, velocity, and displacement responses. Especially when dealing with high-frequency dynamic loads and nonlinear responses, this method exhibits superior performance and stability. Through multi-window time-frequency analysis and wavelet transform, the transient response characteristics were accurately extracted, providing reliable dynamic features for risk assessment.

[0082] According to one aspect of this application, step S43 specifically comprises:

[0083] Step S431: Receive warning level and level stability data; read historical decision scheme set from decision knowledge base; analyze the characteristics of historical schemes using association rule mining algorithm; construct multidimensional decision variable space; define the value range of decision variables based on fuzzy set theory; generate candidate decision schemes using Latin hypercube sampling method; evaluate the feasibility index of each scheme; filter the decision space according to feasibility index; output feasible decision scheme set.

[0084] Step S432: Import the set of feasible decision schemes output in step S431, and read the index calculation parameters from the evaluation index library; construct a benefit evaluation model based on fuzzy hierarchical analysis; calculate the timeliness index of the scheme using the DEA method; estimate the cost of the scheme based on the life cycle theory; incorporate the risk evolution trend into the benefit calculation; determine the index weights using the entropy weight method; calculate the comprehensive evaluation score of each scheme; and output the multi-objective evaluation index values.

[0085] Step S433: Read the multi-objective evaluation index values ​​output in step S432, import resource constraint parameters from the constraint library; construct a resource occupancy model considering time series; use a constraint satisfaction algorithm to verify the feasibility of the scheme; establish a constraint processing model based on fuzzy optimization; calculate the degree of constraint violation; design a penalty function for constraint violation; filter the schemes according to the penalty value; output a set of decision schemes that satisfy the constraints.

[0086] Step S434: Receive the set of decision schemes that satisfy the constraints output in step S433, construct a scheme ranking model based on the PROMETHEE method; use grey relational analysis to calculate the superiority-inferiority relationship between schemes; establish a comprehensive optimization model considering multiple objectives; design a solution algorithm based on simulated annealing; calculate the Pareto optimality of each scheme; select the final scheme according to the optimality ranking; evaluate the robustness index of the schemes; output the optimal decision scheme and optimization index.

[0087] By constructing a multi-objective decision optimization and execution evaluation mechanism, intelligent and optimized early warning decision-making is achieved. This step innovatively introduces a comprehensive evaluation model based on fuzzy hierarchical analysis and DEA methods, combined with constraint satisfaction algorithms for decision optimization. In the scenario of elevator roof load-bearing capacity assessment, this method can comprehensively consider multiple objectives such as benefits, timeliness, and cost to generate the optimal early warning decision scheme. Through the PROMETHEE method and grey relational analysis, the scientific ranking and selection of decision schemes are achieved. This technological innovation not only improves the scientific rigor and reliability of decision-making but also enables continuous optimization and performance improvement of the early warning system. By establishing a robust feedback mechanism, the system can continuously optimize the decision model based on actual execution results, improving the accuracy and adaptability of early warning decisions.

[0088] In the above embodiments, the specific implementation process of the relevant algorithms is as follows:

[0089] The LZ complexity algorithm is: LZ(x) = (c·log_b(n)) / n + λ(t)·H(x); where: LZ(x) is the LZ complexity value; c is the pattern count; n is the sequence length; b is the logarithmic base; H(x) = -Σp_i·log(p_i) is the information entropy term; λ(t) is the adaptive weight; p_i is the probability of pattern i; x is the input sequence; t is the time variable.

[0090] The dynamic time warping algorithm is: DTW(X,Y) = min{Σw_k·d(i_k,j_k)·exp(-α·|i_k-j_k|)}·exp(-β·S(X,Y)); where: DTW(X,Y) is the time warping distance; X,Y are the sequences to be aligned; d(i,j) is the local distance; w_k is the path weight; i_k,j_k are the corresponding point indices; S(X,Y) is the sequence similarity; α,β are control parameters.

[0091] The Huber loss function is: L(δ,r) = {0.5·r 2 / σ 2 (t), |r|≤δ(t); δ(t)·|r| / σ 2 (t)-0.5·δ 2 (t), |r|>δ(t)}·exp(-λ·E(t)); where: L(δ,r) is the loss function value; r is the residual; δ(t) is the adaptive threshold; σ 2 (t) represents the dynamic variance; E(t) represents the system energy; λ represents the decay coefficient; and t represents the time variable.

[0092] The Kalman filtering algorithm is: x-(t+1) = F(t)·x-(t) + K(t)·[y(t)-H(t)·x-(t)]·exp(-α·tr(P(t))); where: x-(t) is the state estimate; F(t) is the state transition matrix; K(t) is the gain matrix; y(t) is the observation vector; H(t) is the observation matrix; P(t) is the error covariance; tr(·) is the matrix trace; and α is the adaptive coefficient.

[0093] The Schmidt orthogonalization method is: q_k = v_k - Σ(β_j·<v_k,q_j> / ||q_j|| 2)·q_j·exp(-λ·κ(Q)); where: q_k is the orthogonalization vector; v_k is the original vector; β_j is the dynamic weight; <·,·> is the inner product; ||·|| is the norm; Q is the orthogonal basis matrix; κ(Q) is the condition number; λ is the stability parameter.

[0094] The matching pursuit algorithm is: MP(x,D) = Σα_k·φ_k·exp(-λ_k·||r_k|| 2 / E_k); where: MP(x,D) is the signal reconstruction result; x is the input signal; D is the overcomplete dictionary; α_k is the adaptive coefficient; φ_k is the selected atom; r_k is the residual of the k-th iteration; E_k is the residual energy; λ_k is the dynamic weight; ||·|| is the norm.

[0095] The mutual information algorithm is: MI(X,Y) = Σp(x,y)·log(p(x,y) / (p(x)·p(y)))·exp(-α·T(X,Y)); where: MI(X,Y) is the mutual information value; p(x,y) is the joint probability; p(x) and p(y) are the marginal probabilities; T(X,Y) is the time delay function; α is the decay coefficient; and X and Y are the input sequences.

[0096] The alternating direction multiplier method is: L(x,y,λ) = f(x) + g(y) + <λ,Ax+By-c> + (ρ / 2)||Ax+By-c|| 2 ·exp(-μ·E(t)); where: L(x,y,λ) is the augmented Lagrangian function; f(x),g(y) are the objective functions; λ is the Lagrange multiplier; ρ is the penalty parameter; A,B are the constraint matrices; c is the constant vector; E(t) is the system energy; μ is the control parameter.

[0097] The Newmark integral format is: u(t+Δt) = u(t) + Δt·v(t) + (Δt) 2 / 2)·[(1-2β)·a(t) + 2β·a(t+Δt)]·exp(-γ·|Δa|); where: u is displacement; v is velocity; a is acceleration; β is integration parameter; Δt is time step; γ is stability parameter; Δa is acceleration increment.

[0098] The Krylov subspace method is: K_m(A,b) = span{b,Ab,A} 2 b,...,A^(m-1)b}·exp(-λ·κ(A_m)); where: K_m is an m-dimensional Krylov subspace; A is the system matrix; b is the initial vector; κ(A_m) is the subspace condition number; λ is the adaptive parameter; and m is the subspace dimension.

[0099] The Gram-Schmidt method is: q*_k = v_k - Σ(w_j·<v_k,q_j> )·q_j·exp(-α·||R||_F); where: q*_k is an orthogonal vector; v_k is the original vector; w_j is the dynamic weight; <·,·> is the inner product; q_j is the orthogonalized vector; R is the upper triangular matrix; ||·||_F is the Frobenius norm; α is the stability parameter.

[0100] The Galerkin method is: (Au,v) + α(t)·(∇u,∇v) + β(t)·(u,v) = (f,v)·exp(-λ·E(u)); where: A is the operator; u,v are the trial functions; α(t),β(t) are the adaptive coefficients; ∇ is the gradient operator; (·,·) is the inner product; f is the source term; E(u) is the energy functional; and λ is the control parameter.

[0101] The Welch method is: P(ω) = (1 / K)·Σ|X_k(ω)| 2 ·W(ω)·exp(-α·V(ω)); where: P(ω) is the power spectrum estimate; X_k(ω) is the Fourier transform of the k-th data segment; W(ω) is the adaptive window function; K is the number of data segments; V(ω) is the frequency variance; α is the smoothing parameter; ω is the frequency variable.

[0102] The association rule mining algorithm is: Conf(A→B) = sup(A∪B) / sup(A)·exp(λ·MI(A,B)); where: Conf(A→B) is the confidence score; sup(·) is the support function; MI(A,B) is the mutual information measure; λ is the weight coefficient; and A and B are itemsets.

[0103] The Latin hypercube sampling method is: X(i,j) = (π(i,j) - U(i,j)) / n·exp(-β·C(X)); where: X(i,j) is the sample matrix; π(i,j) is the permutation matrix; U(i,j) is a uniform random number; n is the number of samples; C(X) is the correlation measure; β is the control parameter; and i,j are indices.

[0104] The DEA method is: E(k) = (Σu_i·y_i) / (Σv_j·x_j)·exp(-λ·D(k)); where: E(k) is the efficiency value of the k-th decision unit; u_i is the output weight; y_i is the output index; v_j is the input weight; x_j is the input index; D(k) is the crowding function; and λ is the penalty coefficient.

[0105] The constraint satisfaction algorithm is: S(x) = min{Σw_i·g_i(x)} 2}·exp(-α·V(x)); where: S(x) is the satisfaction function; w_i is the constraint weight; g_i(x) is the constraint function; V(x) is the violation metric; α is the adaptive parameter; and x is the decision vector.

[0106] Grey relational analysis is defined as: γ(x0,x_i) = (1 / n)·Σ[(min_i min_k|x0(k)-x_i(k)| +ρ·max_i max_k|x0(k)-x_i(k)|) / (|x0(k)-x_i(k)| + ρ·max_i max_k|x0(k)-x_i(k)|)]·exp(-λ·D(x0,x_i)); where: γ(x0,x_i) is the correlation degree; x0 is the reference sequence; x_i is the comparison sequence; ρ is the resolution coefficient; D(x0,x_i) is the distance function; λ is the decay parameter; n is the sequence length; and k is the time index.

[0107] The permutation entropy algorithm is PE(x) = -Σp(π)·ln(p(π))·exp(-λ·C(x)); where: PE(x) is the permutation entropy; p(π) is the probability distribution of the permutation pattern π; C(x) is the sequence complexity; λ is the adaptive parameter; and x is the input sequence.

[0108] The phase unwinding algorithm is: Φ(ω) = ϕ(ω) + 2π·N(ω)·exp(-α·D(ω)); where: Φ(ω) is the unwinding phase; ϕ(ω) is the wrapping phase; N(ω) is an integer function; D(ω) is the discontinuity measure; α is the smoothing coefficient; and ω is the frequency variable.

[0109] The SVD decomposition is: A = U·Σ·V^T·exp(-μ·κ(A)); where: A is the original matrix; U and V are orthogonal matrices; Σ is the singular value matrix; κ(A) is the condition number; and μ is the stability parameter.

[0110] The prediction correction method is: x(t+Δt) = x_p + α(t)·(x_c - x_p)·exp(-β·||r||); where: x_p is the predicted value; x_c is the corrected value; α(t) is the adaptive weight; r is the residual vector; β is the control parameter; and t is the time variable.

[0111] The EMD decomposition is: x(t) = Σc_i(t)·exp(-λ_i·E_i(t)) + r_n(t); where: c_i(t) is the intrinsic mode function; E_i(t) is the modal energy; λ_i is the decay coefficient; r_n(t) is the residual; and t is the time variable.

[0112] Example 2

[0113] S1. Multi-source data acquisition and intelligent preprocessing

[0114] S11. Adaptive Multi-Source Data Acquisition

[0115] Acquire the raw data stream R(t) from multiple sensors, including: load sensor data: L(t); vibration sensor data: V(t); displacement sensor data: D(t); temperature sensor data: T(t);

[0116] 1) Construct a sensor importance scoring function:

[0117] W(i,t) = α·S(i,t) + β·N(i,t) + γ·C(i,t)

[0118] Where: i is the sensor number; S(i,t) is the signal strength coefficient; N(i,t) is the noise level coefficient; C(i,t) is the signal rate of change; α, β, γ are dynamic weighting coefficients.

[0119] 2) Calculate the optimal sampling frequency: f(i,t) = f_base · exp(W(i,t)); where f_base is the reference sampling frequency.

[0120] The synchronized optimized data stream R'(t) is obtained.

[0121] S12. Nonlinear noise suppression

[0122] Read the synchronized optimized data stream R'(t), and then:

[0123] 1) Decompose the signal into multi-dimensional components:

[0124] R'(t) = Σ[H_k(t)] + ε(t)

[0125] Where: H_k(t) is the k-dimensional harmonic component; ε(t) is the residual term.

[0126] 2) Construct the adaptive filtering function:

[0127] F(ω,t) = 1 exp(-λ·|H(ω,t)| 2 / σ 2 (t))

[0128] Where: ω is the frequency component; H(ω,t) is the harmonic amplitude; σ 2 (t) represents the local variance; λ is the adaptive coefficient;

[0129] The denoised data stream R''(t) is obtained.

[0130] S13. Intelligent Anomaly Detection

[0131] Read the denoised data stream R''(t), and then...

[0132] 1) Construct the data manifold space:

[0133] M(t) = {x(t) ∈ R^n | φ(x(t)) = 0}

[0134] Where: φ(x) is the manifold constraint function; n is the data dimension.

[0135] 2) Calculate the topological eigenvectors:

[0136] v(t) = grad(M(t)) / ||grad(M(t))||;

[0137] 3) Anomaly detection criteria:

[0138] A(t) = ||v(t) v̄(t)|| / σ_v(t)

[0139] Where: v̄(t) is the local average eigenvector; σ_v(t) is the standard deviation of the eigenvector;

[0140] Obtain the anomaly-marked data stream R'''(t).

[0141] S14. Data Reconstruction and Compensation

[0142] Read the exception flag data stream R'''(t), then

[0143] 1) Construct a local reconstruction function:

[0144] Φ(t,x) = Σ[w_i(t)·B_i(x)];

[0145] Where: w_i(t) are time-varying weighting coefficients; B_i(x) are basis functions;

[0146] 2) Optimize reconstruction error:

[0147] E(t) = ||R'''(t) Φ(t,x)|| 2 ;

[0148] 3) Update the compensation coefficient: w_i(t+1) = w_i(t) η·∂E / ∂w_i;

[0149] The final processed data stream R_final(t) is obtained;

[0150] S2. Multidimensional Feature Extraction and Load Analysis

[0151] S21. Time-frequency feature extraction

[0152] Read the final data stream R_final(t) output by S1, and then:

[0153] 1) Constructing a generalized time-frequency transform operator:

[0154] Ψ(t,ω) = ∫R_final(τ)·K(t-τ,ω)dτ;

[0155] Where: K(t,ω) is the adaptive kernel function; K(t,ω) = g(t)·exp(jωt + h(t,ω)); g(t) is the time-domain window function; h(t,ω) is the phase compensation function;

[0156] 2) Calculate the time-frequency characteristic matrix: TF(t,ω) = |Ψ(t,ω)| 2 · exp(jφ(t,ω)); where φ(t,ω) represents phase information;

[0157] Obtain the time-frequency characteristic matrix TF_matrix;

[0158] S22. Load Mode Decomposition

[0159] Read the time-frequency feature matrix TF_matrix, and then:

[0160] 1) Construct load mode basis functions:

[0161] B_k(t,ω) = α_k(t)·exp(jβ_k(ω));

[0162] Where: α_k(t) is the time-varying amplitude function; β_k(ω) is the frequency phase function; k is the mode number;

[0163] 2) Optimize the mode decomposition equations:

[0164] TF_matrix = Σ[λ_k·B_k(t,ω)] + ε(t,ω);

[0165] Where: λ_k is the modal weighting coefficient; ε(t,ω) is the residual term;

[0166] 3) Calculate modal correlation:

[0167] C(i,j) =<B_i,B_j> / ||B_i||·||B_j||;This yields the load mode set LM = {(λ_k, B_k)};

[0168] S23. Load Feature Fusion

[0169] Read the load mode set LM;

[0170] 1) Construct a feature importance evaluation function:

[0171] I(k) = μ·E(k) + ν·V(k) + ξ·S(k);

[0172] Where: E(k) is the energy contribution rate; V(k) is the coefficient of variation; S(k) is the stability index; μ, ν, ξ are the weighting coefficients;

[0173] 2) Feature selection criteria:

[0174] F(k) = I(k)·exp(-ρ·C_max(k));

[0175] Where: C_max(k) is the maximum correlation coefficient; ρ is the penalty factor;

[0176] 3) Fusing feature vectors:

[0177] X(t) = Σ[F(k)·λ_k·B_k(t)]; This yields the fused feature vector X(t).

[0178] S24. Construction of Dynamic Load Spectrum

[0179] Read the fused feature vector X(t).

[0180] 1) Construct the load probability density function:

[0181] p(x,t) = Σ[w_i(t)·G(x|μ_i(t),Σ_i(t))];

[0182] Where: G(·) is the generalized Gaussian kernel function; μ_i(t) is the time-varying mean; Σ_i(t) is the covariance matrix; w_i(t) is the mixed weight;

[0183] 2) Calculate the dynamic load spectrum:

[0184] L(x,t) = -∂ln(p(x,t)) / ∂x;

[0185] 3) Construct reliability metrics:

[0186] R(t) = exp(-∫L(x,t)dx);

[0187] We obtained: dynamic load spectrum L(x,t); reliability index R(t);

[0188] S3. Multidimensional Status Assessment and Risk Analysis

[0189] S31. Multiscale state mapping

[0190] Read the dynamic load spectrum L(x,t); reliability index R(t);

[0191] 1) Constructing a multi-scale state space:

[0192] S(t) = {M(t), L(t), D(t)};

[0193] Where: M(t) is the macroscopic state vector (overall structural response); L(t) is the local state vector (stress distribution); D(t) is the microscopic state vector (material properties);

[0194] 2) Define the state transition operator:

[0195] Γ(s1,s2) = exp(-||s1-s2|| 2 / θ(t));

[0196] Where: s1, s2 are state vectors; θ(t) is the adaptive bandwidth function; θ(t) = θ o ·exp(-η·|dR(t) / dt|).

[0197] 3) Calculate the state mapping matrix:

[0198] Q(t) = {q_ij(t)}; q_ij(t) = Γ(s_i,s_j)·ω(t); where ω(t) is the time-varying weight function;

[0199] The state mapping matrix Q(t) is obtained;

[0200] S32. Nonlinear Dynamic Response Analysis

[0201] Read the state mapping matrix Q(t);

[0202] 1) Constructing nonlinear dynamic equations:

[0203] M(t)·x** + C(x,x*,t)·x* + K(x,t)·x = F(t); where: M(t) is the time-varying mass matrix; C(x,x*,t) is the nonlinear damping matrix; K(x,t) is the stiffness matrix; F(t) is the external load vector;

[0204] 2) Propose a new type of solver:

[0205] x(t+Δt) = x(t) + Σ[α_k·Φ_k(x,t)];

[0206] Where: Φ_k is the dynamic basis function; α_k is the optimization coefficient; the solution is based on the variational principle: min J = ||M·x**+ C·x* + K·x F|| 2 ;

[0207] 3) Calculate response characteristics:

[0208] ψ(t) = {A(t), V(t), D(t)}; where: A(t) is the acceleration response spectrum; V(t) is the velocity response spectrum; D(t) is the displacement response spectrum; thus, the dynamic response characteristics ψ(t) are obtained.

[0209] S33. Risk Assessment

[0210] Read the dynamic response characteristics ψ(t);

[0211] 1) Construct a risk assessment indicator system:

[0212] R(t) = {R_s(t), R_d(t), R_f(t)};

[0213] Where: R_s(t) is the static risk indicator; R_d(t) is the dynamic risk indicator; R_f(t) is the fatigue risk indicator;

[0214] 2) Define the comprehensive risk assessment function:

[0215] Ω(t) = ξ1·R_s(t) + ξ2·R_d(t) + ξ3·R_f(t);

[0216] Where: ξᵢ = exp(H(R_i)) / Σ[exp(H(R_j))]; H(·) is the information entropy function;

[0217] 3) Construct the risk evolution equation:

[0218] dΩ / dt = λ(t)·Ω + μ(t)·grad 2 Ω + f(ψ,t);

[0219] Where: λ(t) is the attenuation coefficient; μ(t) is the diffusion coefficient; f(ψ,t) is the source term function;

[0220] The comprehensive risk index Ω(t) and the risk evolution trend dΩ / dt are obtained.

[0221] S34. Reliability Prediction

[0222] Read the comprehensive risk index Ω(t); risk evolution trend dΩ / dt;

[0223] 1) Constructing a state prediction model:

[0224] Z(t+τ) = Λ(Z(t)) + Φ(Ω(t)) + ε(t);

[0225] Where: Z(t) is the system state vector; Λ(·) is the state transition function; Φ(·) is the risk impact function; ε(t) is the random disturbance term;

[0226] 2) Define the reliability evaluation function:

[0227] P(t,τ) = P{Z(t+τ) ∈ S_safe|Z(t),Ω(t)}; where S_safe is the safety region;

[0228] 3) Calculate the failure probability:

[0229] P_f(t,τ) = 1 ∫∫p(z,ω)·I(z,ω)dzdω;

[0230] Where: p(z,ω) is the joint probability density; I(z,ω) is the indicator function;

[0231] The reliability prediction value P(t,τ) and the failure probability P_f(t,τ) are obtained.

[0232] S4. Adaptive Early Warning Decision System

[0233] S41. Construction of Dynamic Early Warning Thresholds

[0234] Read the reliability prediction value P(t,τ); failure probability P_f(t,τ); comprehensive risk index Ω(t);

[0235] 1) Construct a multi-level early warning threshold function:

[0236] Th(k,t) = Th o (k)·exp(-γ(t)·Δr(t));

[0237] Where: k is the warning level (k=1,2,3,4); Th o (k) represents the initial threshold vector; γ(t) is the time-varying adjustment factor; Δr(t) is the rate of change of risk; γ(t) = γ o ·(1-exp(-η·|dΩ / dt|));

[0238] 2) Define the threshold adaptive equation:

[0239] dTh / dt = α(t)·[Th_target Th(k,t)] + β(t)·d 2 Th / dt 2 ;

[0240] Where: α(t) is the asymptotic coefficient; β(t) is the inertia coefficient; Th_target is the target threshold;

[0241] 3) Calculate the confidence level of the threshold:

[0242] C(k,t) = exp(-||Th(k,t) Th*(k,t)|| 2 / σ 2 (t)); where Th*(k,t) is the historical optimal threshold;

[0243] The dynamic early warning threshold set Th(k,t) and threshold confidence level C(k,t) are obtained.

[0244] S42. Warning Level Assessment

[0245] Read the dynamic early warning threshold set Th(k,t); S3 outputs the dynamic response characteristic ψ(t);

[0246] 1) Construct the early warning discrimination function:

[0247] W(t) = Σ[ω_i(t)·φ_i(ψ_i(t),Th_i(k,t))];

[0248] Where: ω_i(t) is the dynamic weight coefficient; φ_i is the discriminant function; ψ_i is the response feature component;

[0249] 2) Define the level conversion rules:

[0250] L(t) = arg max_k{P(k|W(t))};

[0251] Where: P(k|W(t)) = π_k·f_k(W(t)) / Σ[π_j·f_j(W(t))]; π_k is the prior probability; f_k is the conditional density function;

[0252] 3) Calculate the stability index of the grade:

[0253] S(t) = 1 / H(P(k|W(t))) / log(K); where H(·) is the information entropy function.

[0254] The warning level L(t) and the level stability S(t) are obtained.

[0255] S43. Decision Generation and Optimization

[0256] Read the warning level L(t); level stability S(t); risk evolution trend dΩ / dt output by S3;

[0257] 1) Constructing the decision space:

[0258] D(t) = {d_i(t) | i=1,2,...,N};

[0259] Where d_i(t) is the optional decision vector: d_i(t) = [a_i(t), m_i(t), r_i(t)]; a_i(t) is the alarm strategy; m_i(t) is the maintenance suggestion; r_i(t) is the response measure;

[0260] 2) Define the decision optimization objective:

[0261] J(d,t) = λ1·E(d,t) + λ2·T(d,t) + λ3·C(d,t);

[0262] Where: E(d,t) is the benefit function; T(d,t) is the time-efficiency function; C(d,t) is the cost function; λᵢ is the weighting coefficient;

[0263] 3) Solving for the optimal decision:

[0264] d*(t) = arg min_d{J(d,t)}; st G(d,t) ≤ 0; H(d,t) = 0;

[0265] Where: G(d,t) is the inequality constraint; H(d,t) is the equality constraint;

[0266] The optimal decision scheme d*(t) is obtained; the decision optimization index J*(t) is obtained;

[0267] S44. Feedback and Self-Learning

[0268] Read the optimal decision scheme d*(t); the decision optimization index J*(t); and the historical decision database D_hist;

[0269] 1) Construct the decision evaluation function:

[0270] E(t) = ||d*(t) d_real(t)|| 2 Where d_real(t) represents the actual execution result;

[0271] 2) Define knowledge update rules:

[0272] K(t+1) = K(t) + μ(t)·grad E(t);

[0273] Where: K(t) is the knowledge base state; μ(t) is the learning rate; grad E(t) is the gradient term;

[0274] 3) Optimize system parameters:

[0275] θ(t+1) = θ(t) η(t)·∂J* / ∂θ;

[0276] Where: θ(t) is the system parameter vector; η(t) is the optimization step size; J* is the optimal decision index;

[0277] The updated knowledge base K(t+1) and the optimized system parameters θ(t+1) are obtained.

[0278] Example 3, Control Example

[0279] I. According to the SC200 / 200 construction elevator instruction manual (100-meter installation height), the foundation bearing capacity is as follows:

[0280] (1) Total self-weight G = weight of cage + weight of outer cage + total weight of guide rail frame + counterweight + load weight

[0281] =2×2000kg+1480kg+66×150kg+2×1000kg +2×2000kg

[0282] =4000kg+1480kg+9900kg+2000 kg +4000kg

[0283] =21380kg

[0284] =21380kg × 0.0098 =209.6 kN

[0285] (2) Calculation of foundation bearing capacity P:

[0286] P = 2 × G (Considering the effects of dynamic load, self-weight error, and wind load on the foundation, the coefficient n = 2 is taken)

[0287] P = 2 × 209.6 = 419.2 kN

[0288] The foundation bearing capacity is 419.2 kN.

[0289] The self-weight of the foundation concrete is: 5.9 * 3.8 * 0.3 * 25 = 168.15 KN

[0290] Total P = 419.2 + 168.15 = 587.35 kN

[0291] The uniformly distributed load during construction is: 587.35 / (5.9 * 3.8) = 26.2 KN / ㎡;

[0292] II. According to the SC200 / 200 construction elevator instruction manual (100-meter installation height), the foundation bearing capacity is as follows:

[0293] (1) Total self-weight G = weight of cage + weight of outer cage + total weight of guide rail frame + counterweight + load weight

[0294] =2×2000kg+1480kg+132×150kg+2×1000kg +2×2000kg

[0295] =4000kg+1480kg+19800kg+2000 kg +4000kg

[0296] =31280kg

[0297] =31280kg × 0.0098 =306.6 kN

[0298] (2) Calculation of foundation bearing capacity P:

[0299] P = 2 × G (Considering the effects of dynamic load, self-weight error, and wind load on the foundation, the coefficient n = 2 is taken)

[0300] P = 2 × 306.6 = 613.2 kN

[0301] The foundation bearing capacity is 613.2 kN.

[0302] The self-weight of the foundation concrete is: 5.9 * 3.8 * 0.3 * 25 = 168.15 KN

[0303] Total P = 613.2 + 168.15 = 781.35 kN

[0304] The uniformly distributed load during construction is: 587.35 / (5.9 * 3.8) = 34.9 KN / ㎡;

[0305] The basement grid is designed for 8m x 8m, the floor height for 5m, and the slab thickness for 250mm.

[0306] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function, characterized in that, Includes the following steps: Step S1: Collect data streams from multiple sensors and preprocess them to obtain the final processed data stream; Step S2: Receive the final processed data stream, construct the dynamic load spectrum, and establish reliability indicators; Step S3: Receive the dynamic load spectrum and reliability index, and input these data into the macroscopic structural response analysis module, local stress distribution analysis module, and material performance analysis module respectively to calculate the state vectors at three levels; establish the state transition matrix based on the distance relationship between the three levels of state vectors; input the state transition matrix into the dynamic equation solver to calculate the acceleration response spectrum, velocity response spectrum, and displacement response spectrum of the system; Static risk value, dynamic risk value and fatigue risk value are calculated based on response spectrum data, and weight coefficients are determined according to the information entropy of each risk indicator to generate comprehensive risk index and risk evolution trend. Based on the risk evolution trend, predict the future state distribution, calculate the probability of the system entering a failure state, and obtain the reliability prediction value and the failure probability prediction value. Step S4: Receive the reliability prediction value, failure probability prediction value and comprehensive risk index, calculate the initial warning threshold based on historical data, and combine the warning threshold with the risk change rate to construct a dynamic threshold function; The warning discrimination value is calculated based on the dynamic threshold function and dynamic response characteristics. The warning discrimination value is combined with the prior probability distribution to obtain the warning level. A decision space is constructed according to the warning level and the risk evolution trend. The benefit value, timeliness index and cost index of each optional decision scheme are calculated. The optimal decision scheme is obtained through multi-objective optimization. The execution effect of the optimal decision scheme is compared with the expected goal, the evaluation error is calculated, and the knowledge base parameters and system parameters are updated according to the error.

2. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function according to claim 1, characterized in that, Step S1 specifically involves: Step S11: According to the preset reference sampling frequency, acquire load sensor data, vibration sensor data, displacement sensor data and temperature sensor data respectively, and resample the original data stream using the optimal sampling frequency to obtain the synchronized optimized data stream; Step S12: Receive the synchronized optimized data stream, construct an adaptive filtering function, and recombine the processed harmonic components and residual terms to obtain the denoised data stream; Step S13: Import the noise-reduced data stream, determine outliers based on the preset standard deviation threshold, and output the data stream with outlier markers; Step S14: Read the data stream with anomaly markers, perform smooth connection, and finally output the processed data stream.

3. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function as described in claim 1. Its features are, Step S2 specifically involves: Step S21: Import the final processed data stream, calculate the time-frequency energy distribution, and form a time-frequency feature matrix; Step S22: Read the time-frequency feature matrix, construct the time-varying amplitude function and frequency phase function, combine these functions to form the load mode basis function; determine the weight coefficient of each mode; combine the weight coefficient with the corresponding load mode to output the load mode set; Step S23: Receive the load mode set, calculate the energy value, variance and time series stability index of each mode; screen the modes based on the selection criteria, and weight the retained modes according to the importance score to generate a fusion feature vector; Step S24: Import the fused feature vector, calculate the system reliability index, and output the dynamic load spectrum and reliability index.

4. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function as described in claim 1. Its features are, Step S3 specifically involves: Step S31: Import the dynamic load spectrum and reliability index, calculate the overall structural response vector using the finite element model, calculate the local stress distribution vector using the stress concentration factor, and generate the microscopic performance vector by combining the material property curve; calculate the Euclidean distance between adjacent state vectors and construct a state transition function with a time decay term; dynamically adjust the bandwidth parameter of the state transition function according to the rate of change of the reliability index; apply the adjusted state transition function to each state vector to generate a state transition probability matrix; apply a time weight to the state transition probability matrix and output the state mapping matrix. Step S32: Read the state mapping matrix and construct a time-varying mass matrix based on the system mass distribution; The damping matrix is ​​constructed based on the nonlinear relationship between displacement and velocity; the stiffness matrix is ​​calculated by combining material properties and geometric features. Substitute these matrices into the dynamic equations; construct orthogonal dynamic basis functions, and calculate the combination coefficients of the basis functions using the variational principle; solve the dynamic equations through iterative optimization to obtain time-domain response data; perform Fourier transform on the time-domain response data to obtain the acceleration response spectrum, velocity response spectrum, and displacement response spectrum, respectively. Step S33: Receive three types of response spectrum data; calculate the static over-limit probability based on extreme value distribution theory to obtain the static risk index; calculate the dynamic impact effect using the time integral of the response spectrum to obtain the dynamic risk index; calculate the cumulative fatigue damage value based on the cycle number and amplitude distribution of the response spectrum to obtain the fatigue risk index; calculate the time series entropy values ​​of the three risk indices; determine the weighting coefficient of each risk index based on the entropy value; substitute the weighted risk indices into the risk evolution equation; solve the risk evolution equation to obtain the comprehensive risk index and risk evolution trend. Step S34: Import comprehensive risk indicators and risk evolution trends to establish a prediction model describing the system state transition; construct a risk impact function based on the risk indicators; introduce random disturbance terms into the state prediction model to simulate the random fluctuation characteristics of the system; define the safety domain boundary conditions of the system; generate a large number of state samples using the Monte Carlo method; statistically analyze the proportion of samples located within the safety domain to obtain the reliability prediction value. Calculate the probability of the sample crossing the safety domain boundary for the first time to obtain the failure probability prediction value; output the reliability prediction value and the failure probability prediction value.

5. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function as described in claim 1. Its features are, Step S4 specifically involves: Step S41: Import reliability prediction values, failure probability prediction values, and comprehensive risk indicators; read the initial thresholds for each level from the early warning level database; calculate the time derivative of the comprehensive risk indicator to obtain the risk change rate; substitute the risk change rate into the exponential decay function to calculate the threshold adjustment factor; multiply the initial thresholds by the adjustment factor to generate dynamic early warning thresholds. Calculate the mean squared error between the current warning threshold and the historical best threshold to obtain the threshold confidence level; construct constraint equations based on the threshold confidence level and obtain the target threshold through numerical solution; The gradient descent method is used to iteratively optimize the warning threshold until the convergence condition is met, and the dynamic warning threshold set and threshold confidence are output. Step S42: Read the dynamic warning threshold set and dynamic response features, obtain the historical statistical distribution of each response feature from the feature database; calculate the deviation between the current response feature and the warning threshold; calculate the alarm weight coefficient of each feature based on the deviation; substitute the weighted feature deviation into the discriminant function to obtain the warning discrimination value; read the prior probability distribution from the warning level database; calculate the probability density of the warning discrimination value under each level condition. Calculate the posterior probability distribution using Bayes' theorem; select the level with the highest posterior probability as the current warning level. Calculate the entropy value of the posterior probability distribution to obtain the level stability index, and output the warning level and level stability. Step S43: Receive the warning level and level stability, as well as the risk evolution trend, and read the optional alarm strategies, maintenance suggestions and response measures from the decision database; A decision-making benefit evaluation function is constructed, and the early warning level, level stability, and risk evolution trend are substituted into the function to calculate the benefit value; the timeliness index is calculated based on the availability of maintenance resources and scheduling constraints; the cost index is estimated based on the consumption of manpower, material resources, and time; and the benefit value, timeliness index, and cost index are substituted into the objective function. Decision constraints are constructed, including resource limitations and operational procedure requirements; the Lagrange multiplier method is used to solve the constrained optimization problem to obtain the optimal decision scheme and optimization index; Step S44: Import the optimal decision-making scheme and optimization indicators, and obtain real-time feedback data on decision execution from the monitoring system; Calculate the deviation between the actual execution effect and the optimal decision scheme; construct a decision evaluation index and substitute the deviation into the evaluation function to calculate the evaluation score; read system parameters and historical evaluation data from the knowledge base; calculate the knowledge update gradient based on the evaluation score; substitute the gradient information into the parameter update equation; obtain new system parameters through iterative calculation; store the updated system parameters and execution effect data into the knowledge base; output the updated knowledge base and optimized system parameters.

6. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function as described in claim 2. Its features are, Step S11 specifically involves: Step S111: According to the preset initial sampling frequency of the sensors, raw data are collected from the load sensor, vibration sensor, displacement sensor and temperature sensor respectively, and the data stream of each sensor is timestamped and calibrated; the data stream is reordered according to the time sequence relationship of the timestamps. The calibration curves of each sensor are read from the sensor parameter library, and the raw data are converted into physical quantities. Dimensional normalization is performed on each data stream, and the normalized multi-source sensor data stream is output. Step S112: Import the normalized data stream and divide the data into data segments according to a preset time window; calculate the signal variance for each data segment to obtain the signal strength coefficient; calculate the noise level coefficient of the data segment using the multi-scale permutation entropy algorithm; calculate the signal change rate based on the LZ complexity algorithm; multiply these three coefficients by the preset weight coefficients read from the weight database; perform feature space mapping on the weighted coefficients and output the importance score value of each sensor. Step S113: Receive the sensor importance score value and read the benchmark sampling frequency from the sampling parameter library; construct a frequency adjustment equation based on the Sigmoid function; substitute the importance score value into the frequency adjustment equation; introduce a dynamic compensation term, which is adaptively adjusted according to the score difference between adjacent sensors; The optimal sampling frequency for each sensor is obtained through iterative calculation; the sampling frequency is quantized to ensure that it meets the hardware sampling capability constraints, and the optimized sampling frequency for each sensor is output. Step S114: Read the optimized sampling frequency and re-acquire data from each sensor; apply anti-aliasing filtering based on wavelet transform to the newly acquired data; interpolate and align the filtered data according to the minimum time interval; construct a time synchronization evaluation function to calculate the phase difference between data streams; use a dynamic time warping algorithm to eliminate the phase difference; perform linear interpolation on the aligned data to ensure that all data streams have a uniform time step; output the synchronized optimized data stream.

7. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function as described in claim 2. Its features are, Step S14 specifically involves: Step S141: Receive the data stream with anomaly markers and read the smoothing window length parameter from the signal analysis parameter library; divide the data stream into multiple overlapping local time intervals based on the rate of change of the data; calculate the Fourier spectrum of the data in each interval; determine the main frequency components of each interval based on the spectral analysis results; select the basis function combination corresponding to the main frequency components from the basis function library; perform orthogonalization on each basis function to form a complete local basis function set; output the local interval division results and the corresponding basis function set. Step S142: Import the local interval partitioning results and basis function set; read the error evaluation parameters from the data processing parameter library; calculate the projection coefficients of non-abnormal data points in each local interval onto the basis function set; construct a local reconstruction function based on the projection coefficients; substitute the reconstruction function into the original data to calculate the residuals; evaluate the residuals using the Huber loss function; construct an optimization objective function based on the gradient information of the residuals; iteratively optimize the projection coefficients using the L-BFGS algorithm; output the optimized reconstruction function parameters. Step S143: Read the reconstruction function parameters and extract the time location information of outliers from the outlier database; calculate the theoretical value of the outlier location based on the reconstruction function; construct an adaptive weighting function that considers the time distance and amplitude difference between outliers and nearby normal points; apply the weighting function between the theoretical value and the actual value; and use the Kalman filtering algorithm to correct the outliers. Calculate the local consistency index of the corrected data; output the corrected and compensated data sequence. Step S144: Receive the corrected and compensated data sequence, calculate the overlapping part of adjacent local intervals; construct a transition function based on improved Hermite interpolation; apply the transition function to the overlapping intervals for data fusion; calculate the gradient continuity index of the fused region; adaptively adjust the parameters of the transition function according to the continuity index; perform final smoothing processing on the entire data sequence; verify the overall smoothness and local fidelity of the processed data; output the final processed data stream.

8. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function as described in claim 3. Its features are, Step S22 specifically involves: Step S221: Import the time-frequency feature matrix and determine the number of modal basis functions based on the singular value distribution of the matrix; construct a generalized Gabor basis function considering time-varying characteristics; orthogonalize the basis functions using the Schmidt orthogonalization method; calculate the time-frequency localization characteristic index of the basis functions; optimize the basis function parameters based on the characteristic index; obtain the optimal basis function set through iterative calculation; output the optimized modal basis function set. Step S222: Receive the modal basis function set and construct a multi-resolution analysis framework based on wavelet packets; calculate the projection coefficients of the signal and basis functions at each resolution level; perform sparse decomposition using the matching pursuit algorithm; and check the time-frequency consistency of the decomposition results. Based on the consistency test results, the main projection components are selected; the selected projection coefficients are sorted by energy level; and an ordered set of modal projection coefficients is output. Step S223: Read the modal projection coefficient set and construct an instantaneous correlation analysis method based on Hilbert transform; calculate the instantaneous phase difference between each mode; use the mutual information algorithm to evaluate the nonlinear correlation between modes; construct the modal correlation matrix; Calculate the eigenvalue decomposition of the correlation matrix; determine the number of independent modes based on the eigenvalue magnitudes; output the modal correlation assessment results; Step S224: Import the modal correlation evaluation results and adopt the weight initialization method based on information entropy; construct the weight optimization objective function considering modal independence; apply the alternating direction multiplier method to solve the weight optimization problem; calculate the stability index of the optimized weights; and fine-tune the weights according to the stability index. Combine the adjusted weights with the corresponding modes; output the load mode set.

9. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function according to claim 4. Its features are, Step S32 specifically involves: Step S321: Receive the state mapping matrix and read the mass distribution data from the system parameter library; construct the time-varying mass matrix using the Newmark integral format; construct the nonlinear damping matrix based on the coupling relationship between velocity and displacement; read the constitutive relation data from the material parameter library; The geometric stiffness matrix is ​​constructed using a piecewise linearization method; Incorporate the influence of ambient temperature and humidity into stiffness calculation; Correct the singular values ​​of the stiffness matrix; output the mass matrix, damping matrix, and stiffness matrix of the system; Step S322: Read the system matrix and construct dynamic basis functions based on the Krylov subspace method; determine the main components of the basis functions using SVD decomposition; construct generalized basis functions considering higher-order modes; calculate the orthogonality index of the basis functions; orthogonalize the basis functions using the Gram-Schmidt method; evaluate the computational complexity of the basis functions; optimize the number of basis functions according to complexity and accuracy requirements; output the optimized dynamic basis function set. Step S323: Import the dynamic basis function set and construct the displacement interpolation function based on the variational principle; discretize the dynamic equations using the Galerkin method; construct an iterative solution scheme considering nonlinear terms; design an adaptive time step control strategy; and solve the nonlinear equation set using the predictive correction method. Calculate the residual vector at each time step; dynamically adjust the iteration parameters based on the residual magnitude. Output time-domain response data; Step S324: Receive time-domain response data and perform power spectrum estimation using the Welch method; construct a multi-window time-frequency analysis framework; calculate the system's frequency response function; extract transient response features using wavelet transform; resample the response data in the frequency domain; calculate the acceleration spectrum, velocity spectrum, and displacement spectrum respectively; smooth the response spectrum; and output the three types of response spectra.

10. The method for assessing the load-bearing capacity of the roof slab of a passenger / freight elevator with adaptive early warning function according to claim 5. Its features are, Step S43 specifically involves: Step S431: Receive warning level and level stability data, and read the historical decision scheme set from the decision knowledge base; use association rule mining algorithm to analyze the characteristics of historical schemes; Construct a multidimensional decision variable space; define the range of values ​​for decision variables based on fuzzy set theory; generate candidate decision schemes using the Latin hypercube sampling method; evaluate the feasibility indicators of each scheme; and filter the decision space based on the feasibility indicators. Output a set of feasible decision options; Step S432: Import the set of feasible decision schemes and read the indicator calculation parameters from the evaluation indicator library; construct a benefit evaluation model based on fuzzy hierarchical analysis. The DEA method is used to calculate the timeliness index of the scheme; the scheme cost is estimated based on the life cycle theory; the risk evolution trend is incorporated into the benefit calculation; the entropy weight method is used to determine the index weights; the comprehensive evaluation score of each scheme is calculated; and the multi-objective evaluation index values ​​are output. Step S433: Read the multi-objective evaluation index values ​​and import resource constraint parameters from the constraint library; construct a resource occupancy model considering time series; use a constraint satisfaction algorithm to verify the feasibility of the scheme; establish a constraint processing model based on fuzzy optimization; calculate the degree of constraint violation; design a penalty function for constraint violation; filter the schemes according to the penalty value; output a set of decision schemes that satisfy the constraints. Step S434: Receive the set of decision schemes that meet the constraints, construct a scheme ranking model based on the PROMETHEE method; use grey relational analysis to calculate the superiority-inferiority relationship between schemes; establish a comprehensive optimization model considering multiple objectives; design a solution algorithm based on simulated annealing; calculate the Pareto optimality of each scheme; select the final scheme according to the optimality ranking; evaluate the robustness index of the schemes. Output the optimal decision-making scheme and optimization indicators.

Images