A photovoltaic flexible support structure parameter self-adaptive optimization method and system

Abstract

The application provides a photovoltaic flexible support structure parameter adaptive optimization method and system, which comprises the following steps: determining the change range in the whole life cycle of the photovoltaic flexible support corresponding to the geometric parameters, the material parameters and the boundary condition parameters, and converting the change range into a corresponding interval field by using a random field discrete algorithm; generating a corresponding parameter sample set by using a sparse grid sampling algorithm, embedding the upper and lower boundaries of the interval field into a preset basis function of a polynomial as a hard constraint, combining the parameter sample set, and constructing a polynomial proxy model constrained by the interval field; detecting the actual structure response parameters; generating an updated sample set according to the interval deviation and the parameter sample set by using an incremental sparse grid sampling algorithm, and performing online optimization processing on the actual structure response parameters by using the updated sample set to generate optimized structure response parameters. The application can greatly improve the parameter optimization efficiency.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic technology, and in particular to an adaptive optimization method and system for the structural parameters of a flexible photovoltaic support structure. Background Technology

[0002] As the photovoltaic industry rapidly expands into complex terrains such as mountains, tidal flats, and coal mining subsidence areas, flexible photovoltaic supports have become one of the mainstream support forms for large-scale ground-mounted photovoltaic power plants due to their significant advantages such as large span, high land utilization rate, and strong terrain adaptability. To simultaneously ensure structural safety under extreme operating conditions and power generation efficiency during daily operation, dynamic adaptive optimization technology of structural parameters during the operation phase has gradually become a key focus of industry research and development. Among these, active wind and snow resistance control is a core component in improving the performance of the support system throughout its entire life cycle.

[0003] Currently, publicly available adaptive control methods for photovoltaic flexible support systems generally employ threshold control strategies based on simplified linear models. These methods collect environmental and structural parameters such as wind speed, wind direction, cable force, and vibration acceleration by deploying a small number of sensors within the support system. They pre-set one or more fixed thresholds and use a simple logical judgment—"if a parameter exceeds the threshold, a preset action is executed"—to adjust key parameters such as cable net pretension and component tilt angle. Due to its low implementation difficulty, minimal computational load, and controllable deployment costs, this control method has been initially applied in existing intelligent flexible support demonstration projects.

[0004] However, flexible photovoltaic supports are essentially large-span prestressed cable-net structures, exhibiting significant geometric nonlinearity, material nonlinearity, and fluid-structure interaction effects. Their structural response displays a strongly nonlinear relationship with environmental loads. Existing simplified linear models cannot accurately describe this complex nonlinear mapping. Under extreme conditions such as strong winds and heavy snow, the structural response prediction error can reach over 40%, leading to a severe deviation between control actions and actual requirements. Fixed threshold control cannot adapt to the dynamic changes in structural characteristics under different operating conditions, easily resulting in control lag or over-control. This can cause significant losses in power generation efficiency or, in severe cases, structural instability accidents. Summary of the Invention

[0005] Based on this, the purpose of this invention is to provide an adaptive optimization method and system for the structural parameters of photovoltaic flexible support structures, so as to solve the problem that the existing technology cannot accurately describe the complex nonlinear mapping relationship of photovoltaic flexible support structures, and that the structural response error is large under extreme conditions such as strong winds and heavy snow, resulting in the control action deviating from the actual needs and the parameter optimization efficiency being low.

[0006] The first aspect of the present invention proposes: An adaptive optimization method for the structural parameters of a photovoltaic flexible support structure, wherein the method includes: The structural parameters of the photovoltaic flexible support are divided into geometric parameters, material parameters, and boundary condition parameters. For the geometric parameters, material parameters, and boundary condition parameters, the variation range of the photovoltaic flexible support throughout its entire life cycle is determined. The variation range is then converted into a corresponding interval field using a random field discretization algorithm. Based on the interval field, a sparse grid sampling algorithm is used to generate a corresponding parameter sample set, and the upper and lower boundaries of the interval field are embedded as hard constraints into the preset basis functions of the polynomial. Combined with the parameter sample set, a polynomial surrogate model with interval field constraints is constructed. Based on the polynomial surrogate model, the interval sensitivity corresponding to the geometric parameters, the material parameters and the boundary condition parameters is calculated respectively, and the corresponding actual structural response parameters are detected during the actual operation of the photovoltaic flexible support. The actual structural response parameters are compared with the interval sensitivity to calculate the corresponding interval deviation. When the interval deviation exceeds a preset deviation threshold, an incremental sparse grid sampling algorithm is used to generate a corresponding updated sample set based on the interval deviation and the parameter sample set. The actual structural response parameters are then optimized online using the updated sample set to generate corresponding optimized structural response parameters.

[0007] The beneficial effects of this invention are as follows: This solution classifies and defines the structural parameters of photovoltaic flexible supports and the range of parameter changes throughout their entire life cycle. It constructs interval fields using a random field discretization algorithm, generates parameter sample sets based on sparse grid sampling, and uses the interval field boundaries as hard constraints to construct a multinomial surrogate model. This accurately characterizes the complex nonlinear mapping relationship of photovoltaic flexible supports, effectively overcoming the shortcomings of inaccurate mapping descriptions in existing technologies and significantly reducing structural response errors under extreme conditions such as strong winds and heavy snow. Simultaneously, by calculating the interval sensitivity of various parameters and comparing the deviation between the actual structural response parameters and the interval sensitivity in real time, incremental sparse grid sampling is used to update the sample set and complete online optimization when the deviation exceeds a threshold. This effectively avoids the problem of control actions deviating from actual needs, significantly improves the optimization efficiency of photovoltaic flexible support structural parameters, and ensures the accuracy and stability of the support's operation and control under extreme conditions.

[0008] Furthermore, the step of converting the range of variation into a corresponding interval field using a random field discretization algorithm includes: Based on the historical operating data of the photovoltaic flexible support, a Markov transition matrix of the structural parameters changing over time is constructed, and combined with a preset spatial covariance function, the spatiotemporal joint prior distribution of the structural parameters is constructed. The structure of the photovoltaic flexible support is divided into several discrete units by the random field discretization algorithm, and the parameter values ​​of each discrete unit are used as random variables of the conditional random field. The spatiotemporal joint prior distribution is used as the potential function of the conditional random field, and the belief propagation algorithm is used to solve the conditional random field to obtain the marginal probability distribution of the parameter values ​​of each discrete unit. The confidence intervals of each discrete unit are calculated based on the marginal probability distribution, and the confidence intervals are spatially spliced ​​together to generate the interval field.

[0009] Furthermore, the step of spatially stitching the various confidence intervals to generate the corresponding interval field includes: Each discrete unit is used as a graph node, the physical connection relationship between each discrete unit is used as a graph edge, and the weight of the corresponding graph edge is determined according to the coupling coefficient between adjacent discrete units to construct the corresponding uncertainty propagation graph. Each confidence interval is assigned to the interior of each node in the uncertainty propagation graph, and the attention coefficient of each node to all its neighboring nodes is calculated using a graph attention network. Based on the attention coefficient and the edge weight of the node, the confidence interval of each node is weighted and propagated to its neighboring nodes to generate the corresponding spatial interval. Based on the Markov transition matrix, the actual parameter intervals of each discrete unit are calculated, and the spatial intervals and the actual parameter intervals are weighted and fused to generate the interval field.

[0010] Furthermore, the step of embedding the upper and lower boundaries of the interval field as hard constraints into the preset basis functions of the polynomial, and constructing a polynomial surrogate model of the interval field constraints in conjunction with the parameter sample set, includes: The interval field of each discrete unit is decomposed into a central field and a radius field, and the spatial distribution characteristics of the central field and the radius field are detected respectively. The central field is the arithmetic mean of the upper and lower bounds of the interval field, and the radius field is half of the difference between the upper and lower bounds of the interval field. Based on the spatial distribution characteristics, a central field basis function is constructed according to the central field, and a radius field basis function is constructed according to the radius field. The central field basis function and the radius field basis function are then combined by Cartesian product to obtain a set of polynomial basis functions with center-radius separation. Using the parameter sample set as training data, the prediction residuals of the central field and the radius field are calculated respectively. With all output values ​​of the radius field being greater than or equal to zero as hard constraints, the residuals are input into the polynomial basis function set to construct the polynomial surrogate model.

[0011] Furthermore, the step of calculating the interval sensitivities corresponding to the geometric parameters, material parameters, and boundary condition parameters based on the polynomial surrogate model includes: The first-order partial derivatives of the geometric parameters, material parameters, and boundary condition parameters are solved using the polynomial proxy model to obtain the partial derivative polynomial operators corresponding to each parameter type. The parameter space correlation matrix between each discrete unit is constructed based on the preset spatial covariance function, and the partial derivative polynomial operator is multiplied with the parameter space correlation matrix to generate the corresponding extended partial derivative operator. The range of values ​​of the extended partial derivative operator in its corresponding parameter interval is calculated using interval arithmetic method to obtain the upper and lower bounds of the interval sensitivity of each discrete unit. The upper and lower bounds of the interval sensitivity are then integrated to generate the corresponding interval sensitivity.

[0012] Furthermore, the step of performing online optimization processing on the actual structural response parameters using the updated sample set to generate corresponding optimized structural response parameters includes: The partial derivatives of the interval deviation with respect to the geometric parameters, the material parameters, and the boundary condition parameters are calculated to obtain the deviation gradient vector. The sampling step size of each type of parameter is determined according to the magnitude of each deviation gradient vector. The updated sample set is generated according to the sampling step size by the incremental sparse grid sampling algorithm. The updated sample set and the parameter sample set are concatenated to generate a concatenated sample set, and the polynomial proxy model is incrementally updated based on the concatenated sample set. On the updated polynomial surrogate model, with the interval fields of various parameters as constraints, a line search is performed in the opposite direction of the deviation gradient vector to determine the optimal step size. The parameter values ​​of various parameters are updated according to the optimal step size, and the updated parameter values ​​are substituted into the updated polynomial surrogate model to output the optimized structural response parameters.

[0013] Furthermore, the step of substituting the updated parameter values ​​into the updated polynomial surrogate model to output the optimized structural response parameters includes: Based on the interval field hard constraint boundaries corresponding to each type of parameter, the updated parameter values ​​are subjected to interval folding mapping to obtain a set of mapped parameters, wherein all constraint parameter values ​​are within the range of the interval field after the random field is discretized; The incremental correction coefficients and original intrinsic coefficients of the updated polynomial surrogate model are detected, and a dual-coefficient coupling calculation matrix is ​​constructed accordingly. The mapping parameter set is then input into the dual-coefficient coupling calculation matrix for simultaneous solution processing. Based on the hierarchical matching rules of sparse grid sampling, the simultaneous solution results are subjected to sampling hierarchical calibration processing to obtain the corresponding optimized structural response parameters.

[0014] The second aspect of the present invention proposes: An adaptive optimization system for the structural parameters of a photovoltaic flexible support structure, wherein the system comprises: The determination module is used to divide the structural parameters of the photovoltaic flexible support into geometric parameters, material parameters and boundary condition parameters, and for the geometric parameters, the material parameters and the boundary condition parameters, determine the variation range of the photovoltaic flexible support throughout its entire life cycle, and use a random field discretization algorithm to convert the variation range into the corresponding interval field; The generation module is used to generate a corresponding parameter sample set based on the interval field using a sparse grid sampling algorithm, and to embed the upper and lower boundaries of the interval field as hard constraints into the preset basis functions of the polynomial. Combined with the parameter sample set, a polynomial surrogate model of the interval field constraint is constructed accordingly. The calculation module is used to calculate the interval sensitivity corresponding to the geometric parameters, the material parameters and the boundary condition parameters based on the polynomial surrogate model, and to detect the corresponding actual structural response parameters during the actual operation of the photovoltaic flexible support. The optimization module is used to compare the actual structural response parameters with the interval sensitivity to calculate the corresponding interval deviation. When the interval deviation exceeds a preset deviation threshold, an incremental sparse grid sampling algorithm is used to generate a corresponding updated sample set based on the interval deviation and the parameter sample set. The updated sample set is then used to perform online optimization processing on the actual structural response parameters to generate corresponding optimized structural response parameters.

[0015] Furthermore, the determining module is specifically used for: Based on the historical operating data of the photovoltaic flexible support, a Markov transition matrix of the structural parameters changing over time is constructed, and combined with a preset spatial covariance function, the spatiotemporal joint prior distribution of the structural parameters is constructed. The structure of the photovoltaic flexible support is divided into several discrete units by the random field discretization algorithm, and the parameter values ​​of each discrete unit are used as random variables of the conditional random field. The spatiotemporal joint prior distribution is used as the potential function of the conditional random field, and the belief propagation algorithm is used to solve the conditional random field to obtain the marginal probability distribution of the parameter values ​​of each discrete unit. The confidence intervals of each discrete unit are calculated based on the marginal probability distribution, and the confidence intervals are spatially spliced ​​together to generate the interval field.

[0016] Furthermore, the determining module is specifically used for: Each discrete unit is used as a graph node, the physical connection relationship between each discrete unit is used as a graph edge, and the weight of the corresponding graph edge is determined according to the coupling coefficient between adjacent discrete units to construct the corresponding uncertainty propagation graph. Each confidence interval is assigned to the interior of each node in the uncertainty propagation graph, and the attention coefficient of each node to all its neighboring nodes is calculated using a graph attention network. Based on the attention coefficient and the edge weight of the node, the confidence interval of each node is weighted and propagated to its neighboring nodes to generate the corresponding spatial interval. Based on the Markov transition matrix, the actual parameter intervals of each discrete unit are calculated, and the spatial intervals and the actual parameter intervals are weighted and fused to generate the interval field.

[0017] Furthermore, the generation module is specifically used for: The interval field of each discrete unit is decomposed into a central field and a radius field, and the spatial distribution characteristics of the central field and the radius field are detected respectively. The central field is the arithmetic mean of the upper and lower bounds of the interval field, and the radius field is half of the difference between the upper and lower bounds of the interval field. Based on the spatial distribution characteristics, a central field basis function is constructed according to the central field, and a radius field basis function is constructed according to the radius field. The central field basis function and the radius field basis function are then combined by Cartesian product to obtain a set of polynomial basis functions with center-radius separation. Using the parameter sample set as training data, the prediction residuals of the central field and the radius field are calculated respectively. With all output values ​​of the radius field being greater than or equal to zero as hard constraints, the residuals are input into the polynomial basis function set to construct the polynomial surrogate model.

[0018] Furthermore, the calculation module is specifically used for: The first-order partial derivatives of the geometric parameters, material parameters, and boundary condition parameters are solved using the polynomial proxy model to obtain the partial derivative polynomial operators corresponding to each parameter type. The parameter space correlation matrix between each discrete unit is constructed based on the preset spatial covariance function, and the partial derivative polynomial operator is multiplied with the parameter space correlation matrix to generate the corresponding extended partial derivative operator. The range of values ​​of the extended partial derivative operator in its corresponding parameter interval is calculated using interval arithmetic method to obtain the upper and lower bounds of the interval sensitivity of each discrete unit. The upper and lower bounds of the interval sensitivity are then integrated to generate the corresponding interval sensitivity.

[0019] Furthermore, the optimization module is specifically used for: The partial derivatives of the interval deviation with respect to the geometric parameters, the material parameters, and the boundary condition parameters are calculated to obtain the deviation gradient vector. The sampling step size of each type of parameter is determined according to the magnitude of each deviation gradient vector. The updated sample set is generated according to the sampling step size by the incremental sparse grid sampling algorithm. The updated sample set and the parameter sample set are concatenated to generate a concatenated sample set, and the polynomial proxy model is incrementally updated based on the concatenated sample set. On the updated polynomial surrogate model, with the interval fields of various parameters as constraints, a line search is performed in the opposite direction of the deviation gradient vector to determine the optimal step size. The parameter values ​​of various parameters are updated according to the optimal step size, and the updated parameter values ​​are substituted into the updated polynomial surrogate model to output the optimized structural response parameters.

[0020] Furthermore, the optimization module is specifically used for: Based on the interval field hard constraint boundaries corresponding to each type of parameter, the updated parameter values ​​are subjected to interval folding mapping to obtain a set of mapped parameters, wherein all constraint parameter values ​​are within the range of the interval field after the random field is discretized; The incremental correction coefficients and original intrinsic coefficients of the updated polynomial surrogate model are detected, and a dual-coefficient coupling calculation matrix is ​​constructed accordingly. The mapping parameter set is then input into the dual-coefficient coupling calculation matrix for simultaneous solution processing. Based on the hierarchical matching rules of sparse grid sampling, the simultaneous solution results are subjected to sampling hierarchical calibration processing to obtain the corresponding optimized structural response parameters.

[0021] The third aspect of the present invention proposes: A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the adaptive optimization method for photovoltaic flexible support structure parameters as described above.

[0022] The fourth aspect of the present invention proposes: A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the adaptive optimization method for photovoltaic flexible support structure parameters as described above.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] Figure 1 A flowchart of the adaptive optimization method for photovoltaic flexible support structure parameters provided in the first embodiment of the present invention; Figure 2 The structural block diagram of the photovoltaic flexible support structure parameter adaptive optimization system provided in the third embodiment of the present invention is shown.

[0025] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0026] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0027] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] Please see Figure 1The diagram shows the adaptive optimization method for photovoltaic flexible support structure parameters provided in the first embodiment of the present invention. The adaptive optimization method for photovoltaic flexible support structure parameters provided in this embodiment can update the sample set by incremental sparse grid sampling and complete online optimization when the deviation exceeds the threshold. This effectively avoids the problem of deviation between control actions and actual needs, significantly improves the optimization efficiency of photovoltaic flexible support structure parameters, and ensures the accuracy and stability of support operation and control under extreme working conditions.

[0030] Specifically, this embodiment provides: An adaptive optimization method for the structural parameters of a photovoltaic flexible support structure, wherein the method includes: Step S10: Divide the structural parameters of the photovoltaic flexible support into geometric parameters, material parameters and boundary condition parameters, and determine the variation range of the geometric parameters, material parameters and boundary condition parameters throughout the entire life cycle of the photovoltaic flexible support, and use a random field discretization algorithm to convert the variation range into the corresponding interval field. It is important to note that the structural response of photovoltaic flexible supports is determined by three core parameters: geometric parameters, including cable length, span-to-span ratio, support height, and node spacing, which are affected by construction errors and cable elongation under long-term loads; material parameters, including cable elastic modulus, yield strength, membrane tensile strength, and connector stiffness, which gradually deteriorate with UV aging, temperature cycling, and fatigue damage; and boundary condition parameters, including foundation settlement, support stiffness, and anchoring force, which are affected by soil consolidation, groundwater erosion, and cumulative wind vibration. Traditional methods treat these parameters as fixed values, using a single safety factor to cover all uncertainties, leading to conservative or risky designs. This step analyzes the variation patterns of these three types of parameters over a 25-year lifespan to determine their reasonable range of variation. A random field discretization algorithm is then used to expand the parameter variation range from a single numerical interval to an interval field considering spatial correlation and temporal evolution characteristics. Each parameter at a spatial location corresponds to a confidence interval, which more realistically reflects the actual distribution characteristics of the structural parameters, laying the foundation for subsequent uncertainty analysis and optimization.

[0031] Step S20: Based on the interval field, a sparse grid sampling algorithm is used to generate a corresponding parameter sample set, and the upper and lower boundaries of the interval field are used as hard constraints and embedded into the preset basis functions of the polynomial. Combined with the parameter sample set, a polynomial surrogate model of the interval field constraint is constructed accordingly. It should be noted that nonlinear finite element simulations of photovoltaic flexible supports can take hours or even days, which cannot support the large number of parameter iterations required for online optimization. This step employs a sparse mesh sampling algorithm to generate a parameter sample set. Compared to traditional Monte Carlo sampling, sparse mesh can reduce the number of samples by more than an order of magnitude while maintaining the same accuracy, significantly reducing sampling costs. Simultaneously, it innovatively embeds the upper and lower boundaries of the interval field as hard constraints into the design of the polynomial basis function, ensuring that the output of the surrogate model inherently satisfies the parameter range constraints and avoids predictions exceeding physical possibilities. The polynomial surrogate model trained based on the parameter sample set can predict the structural response under arbitrary parameter combinations within milliseconds, improving computational efficiency by tens of thousands of times compared to finite element simulation, thus enabling online adaptive optimization.

[0032] Step S30: Based on the polynomial surrogate model, calculate the interval sensitivity corresponding to the geometric parameters, the material parameters and the boundary condition parameters respectively, and detect the corresponding actual structural response parameters during the actual operation of the photovoltaic flexible support. It's important to note that interval sensitivity analysis answers the core question of "which parameter's change has the greatest impact on structural safety," helping maintenance personnel focus limited resources on the most critical parameters. This step, based on a trained polynomial surrogate model, calculates the interval sensitivity of each of the three types of parameters, obtaining not only the average sensitivity but also the uncertainty range of the sensitivity, comprehensively assessing the degree of parameter impact. Simultaneously, through devices deployed on the support structure, such as strain sensors, cable force sensors, displacement sensors, and tilt sensors, the actual response parameters of the structure are collected in real time, including cable force, nodal displacement, structural vibration frequency, and support reaction force, to understand the true operating state of the structure.

[0033] Step S40: Compare the actual structural response parameters with the interval sensitivity to calculate the corresponding interval deviation. When the interval deviation exceeds a preset deviation threshold, use an incremental sparse grid sampling algorithm to generate a corresponding updated sample set based on the interval deviation and the parameter sample set. Then, use the updated sample set to perform online optimization processing on the actual structural response parameters to generate corresponding optimized structural response parameters.

[0034] It should be noted that when the actual structural response deviates significantly from the range predicted by the surrogate model, it indicates that the structural parameters have deteriorated beyond expectations, or the accuracy of the surrogate model is no longer sufficient. In this case, this step employs an incremental sparse grid sampling algorithm, supplementing only a small number of samples in the parameter regions with large deviations to generate an updated sample set, without requiring global resampling. Then, based on the updated sample set, the surrogate model is incrementally updated, and the structural parameters are adjusted using a gradient optimization algorithm to bring the structural response back to a safe and reasonable range. The entire optimization process is completed online without downtime, adapting in real-time to dynamic changes in structural parameters and achieving adaptive optimization throughout its entire lifecycle.

[0035] Second Embodiment Furthermore, the step of converting the range of variation into a corresponding interval field using a random field discretization algorithm includes: Based on the historical operating data of the photovoltaic flexible support, a Markov transition matrix of the structural parameters changing over time is constructed, and combined with a preset spatial covariance function, the spatiotemporal joint prior distribution of the structural parameters is constructed. The structure of the photovoltaic flexible support is divided into several discrete units by the random field discretization algorithm, and the parameter values ​​of each discrete unit are used as random variables of the conditional random field. The spatiotemporal joint prior distribution is used as the potential function of the conditional random field, and the belief propagation algorithm is used to solve the conditional random field to obtain the marginal probability distribution of the parameter values ​​of each discrete unit. The confidence intervals of each discrete unit are calculated based on the marginal probability distribution, and the confidence intervals are spatially spliced ​​together to generate the interval field.

[0036] It is important to note that the changes in the structural parameters of photovoltaic flexible supports are a spatiotemporally coupled process: in the time dimension, processes such as material aging and foundation settlement exhibit Markov properties, meaning that the future parameter state depends only on the current state and is independent of the past; in the spatial dimension, the parameters of adjacent units are correlated. For example, the cable force at adjacent nodes on the same cable will not change abruptly, and the settlement of adjacent foundations is also correlated. This step, based on a large amount of historical operating data from photovoltaic power plants, statistically analyzes the probability of changes in different parameters over different time intervals, constructs a Markov transition matrix, and describes the evolution of parameters over time. Simultaneously, a suitable spatial covariance function (such as the Gaussian covariance function or the exponential covariance function) is selected to describe the strength and range of spatial correlation of the parameters. By combining the temporal Markov property and spatial correlation, a spatiotemporal joint prior distribution of the structural parameters is constructed, providing a probabilistic basis for subsequent random field discretization.

[0037] This step first divides the entire photovoltaic flexible support structure into several discrete units based on its structural characteristics and computational accuracy requirements. For example, each cable segment, each node, and each foundation section is treated as an independent discrete unit. Then, a conditional random field (CRF) model is constructed, treating the parameter values ​​of each discrete unit as random variables and the spatiotemporal joint prior distribution as the model's potential function. Sensor-measured data is also incorporated as observation conditions. Finally, an efficient belief propagation algorithm is used to solve the CRF, obtaining the marginal probability distribution of the parameter values ​​for each discrete unit. The marginal probability distribution comprehensively describes the uncertainty of the unit's parameters, containing richer probabilistic information than a single numerical range, and can more accurately assess the structure's reliability.

[0038] This step selects the commonly used 95% confidence interval in engineering, meaning that 95% of the parameter values ​​will fall within this interval. This covers the vast majority of possible parameter values, ensuring structural safety, without being overly conservative due to excessively wide intervals. After obtaining the confidence interval for each discrete element, spatial stitching is performed to eliminate interval discontinuities at element boundaries, making the interval field continuous and smooth throughout the entire structural space. The final generated interval field is a three-dimensional spatiotemporal field, where each spatial location corresponds to a parameter confidence interval at each time step, fully characterizing the spatiotemporal uncertainty of the structural parameters throughout their entire lifespan.

[0039] Furthermore, the step of spatially stitching the various confidence intervals to generate the corresponding interval field includes: Each discrete unit is used as a graph node, the physical connection relationship between each discrete unit is used as a graph edge, and the weight of the corresponding graph edge is determined according to the coupling coefficient between adjacent discrete units to construct the corresponding uncertainty propagation graph. Each confidence interval is assigned to the interior of each node in the uncertainty propagation graph, and the attention coefficient of each node to all its neighboring nodes is calculated using a graph attention network. Based on the attention coefficient and the edge weight of the node, the confidence interval of each node is weighted and propagated to its neighboring nodes to generate the corresponding spatial interval. Based on the Markov transition matrix, the actual parameter intervals of each discrete unit are calculated, and the spatial intervals and the actual parameter intervals are weighted and fused to generate the interval field.

[0040] It should be noted that the photovoltaic flexible support is a complex structural system composed of cables, rods, nodes, and foundations. Adjacent units interact through physical connections, and the parameter uncertainty of one unit propagates to adjacent units through these connections. This step abstracts each discrete unit as a graph node and the physical connections between units as graph edges. Then, based on structural mechanics principles, the coupling coefficient between adjacent units is calculated. The larger the coupling coefficient, the stronger the interaction between the two units, and the greater the impact of parameter changes in one unit on the other. The coupling coefficient is used as the weight of the graph edges to construct an uncertainty propagation graph. This graph clearly reflects the uncertainty propagation path and intensity within the structure, providing a foundation for subsequent spatial weighted propagation.

[0041] Traditional spatial stitching methods use fixed average weights, which cannot distinguish the differences in influence between adjacent nodes. This step introduces a graph attention network, which can automatically learn the attention coefficient of each node to its adjacent nodes. The larger the attention coefficient, the greater the influence of the adjacent node on the current node. For example, the attention coefficient of the node of the main load-bearing cable to the adjacent secondary cable node will be much greater than the attention coefficient of the secondary cable node to the main cable node. Then, by combining the attention coefficient with the coupling weight of the graph edges, the confidence interval of each node is propagated to its adjacent nodes in a weighted manner, so that the confidence intervals of adjacent cells merge with each other, eliminating discontinuities at cell boundaries and generating continuous and smooth spatial intervals.

[0042] The spatial intervals obtained from spatially weighted propagation primarily reflect the spatial correlation of parameters, while the actual parameter intervals calculated based on the Markov transition matrix primarily reflect the evolutionary characteristics of parameters over time. This step integrates the two with weights, dynamically adjusting the weights according to the spatiotemporal variation characteristics of the parameters: for parameters with strong spatial correlation (such as the elastic modulus of the cable), the spatial intervals are assigned higher weights; for parameters with significant temporal evolution (such as foundation settlement), the actual parameter intervals are assigned higher weights. Through spatiotemporal weighted fusion, the final generated interval field simultaneously considers the spatial correlation and temporal evolution characteristics of the parameters, and can more accurately reflect the true uncertainty of structural parameters at any time and any location.

[0043] Furthermore, the step of embedding the upper and lower boundaries of the interval field as hard constraints into the preset basis functions of the polynomial, and constructing a polynomial surrogate model of the interval field constraints in conjunction with the parameter sample set, includes: The interval field of each discrete unit is decomposed into a central field and a radius field, and the spatial distribution characteristics of the central field and the radius field are detected respectively. The central field is the arithmetic mean of the upper and lower bounds of the interval field, and the radius field is half of the difference between the upper and lower bounds of the interval field. Based on the spatial distribution characteristics, a central field basis function is constructed according to the central field, and a radius field basis function is constructed according to the radius field. The central field basis function and the radius field basis function are then combined by Cartesian product to obtain a set of polynomial basis functions with center-radius separation. Using the parameter sample set as training data, the prediction residuals of the central field and the radius field are calculated respectively. With all output values ​​of the radius field being greater than or equal to zero as hard constraints, the residuals are input into the polynomial basis function set to construct the polynomial surrogate model.

[0044] It should be noted that each parameter interval in the interval field can be represented as [center value - radius, center value + radius]. The center value represents the nominal value of the parameter, reflecting its overall level; the radius represents the uncertainty range of the parameter, reflecting its fluctuation. The center field and radius field have different spatial distribution characteristics and variation patterns: the center field is usually related to the design values ​​of the structure, and its spatial distribution is relatively smooth; the radius field is usually related to construction errors and material dispersion, and its spatial distribution may exhibit local abrupt changes. Separating the two for modeling allows for the capture of their respective characteristics, significantly improving the prediction accuracy of the surrogate model.

[0045] This step selects appropriate polynomial order and type based on the spatial distribution characteristics of the central and radial fields: for a central field with a smooth spatial distribution, lower-order Legendre polynomials are used as basis functions; for a radial field with drastic spatial distribution changes, higher-order Chebyshev polynomials are used as basis functions. Then, the central field basis functions and the radial field basis functions are combined using a Cartesian product to obtain a set of polynomial basis functions with separated center and radius values. This basis function design can simultaneously model the changes in both the central value and the radius, maintaining their independence and avoiding mutual interference.

[0046] The radius field represents the half-width of the parameter interval, and its physical meaning dictates that the radius value must be greater than or equal to zero; otherwise, the interval becomes meaningless. Traditional surrogate models do not consider this constraint and may output negative radius values, rendering the results invalid. This step embeds "radius field output ≥ 0" as a hard constraint into the training process of the polynomial basis functions. A constraint optimization algorithm is used to solve for the model coefficients, ensuring that all outputs of the surrogate model satisfy this physical constraint. Simultaneously, using the parameter sample set as training data, the prediction residuals of the central field and radius field are minimized to optimize the model coefficients, ultimately resulting in a high-precision, physically-constrained interval-field-constrained polynomial surrogate model.

[0047] Furthermore, the step of calculating the interval sensitivities corresponding to the geometric parameters, material parameters, and boundary condition parameters based on the polynomial surrogate model includes: The first-order partial derivatives of the geometric parameters, material parameters, and boundary condition parameters are solved using the polynomial proxy model to obtain the partial derivative polynomial operators corresponding to each parameter type. The parameter space correlation matrix between each discrete unit is constructed based on the preset spatial covariance function, and the partial derivative polynomial operator is multiplied with the parameter space correlation matrix to generate the corresponding extended partial derivative operator. The range of values ​​of the extended partial derivative operator in its corresponding parameter interval is calculated using interval arithmetic method to obtain the upper and lower bounds of the interval sensitivity of each discrete unit. The upper and lower bounds of the interval sensitivity are then integrated to generate the corresponding interval sensitivity.

[0048] It's important to note that the polynomial surrogate model possesses an analytical expression, allowing for direct differentiation, a significant advantage over black-box models (such as neural networks). This step calculates the first-order partial derivatives of the polynomial surrogate model with respect to geometric parameters, material parameters, and boundary condition parameters, yielding the corresponding partial derivative polynomial operators. The physical meaning of the partial derivative operators is: how much the structural response changes when a certain parameter undergoes a unit change. The larger the absolute value of the partial derivative, the more significant the influence of that parameter on the structural response.

[0049] Traditional sensitivity analysis assumes that the parameters of each element are independent. However, in reality, the parameters of adjacent elements exhibit significant spatial correlation; a change in the parameter of one element will lead to changes in the parameters of adjacent elements. This step generates a parameter spatial correlation matrix based on the previously constructed spatial covariance function, where the elements represent the correlation coefficients between the parameters of two elements. Then, the partial derivative polynomial operator is multiplied with the spatial correlation matrix to obtain the extended partial derivative operator. The extended partial derivative operator considers not only the influence of a single parameter but also the cascading effects of changes in the parameters of adjacent elements, thus more accurately reflecting the true extent to which parameters influence the structural response.

[0050] The extended partial derivative operator is a polynomial function whose input parameters vary within an interval field, thus its output partial derivatives also vary within a certain range. This step uses interval arithmetic to calculate the maximum and minimum values ​​of the extended partial derivative operator within the corresponding parameter intervals, obtaining the upper and lower bounds of the interval sensitivity. These bounds comprehensively reflect the uncertainty of the parameter's influence: the larger the difference between the upper and lower bounds, the higher the uncertainty of the parameter's sensitivity, and the more difficult it is to predict its impact on the structural response. By integrating the upper and lower bounds of the interval sensitivity for all discrete units, a multi-parameter interval sensitivity field for the entire structure is obtained, providing a comprehensive basis for subsequent optimization and adjustment.

[0051] Furthermore, the step of performing online optimization processing on the actual structural response parameters using the updated sample set to generate corresponding optimized structural response parameters includes: The partial derivatives of the interval deviation with respect to the geometric parameters, the material parameters, and the boundary condition parameters are calculated to obtain the deviation gradient vector. The sampling step size of each type of parameter is determined according to the magnitude of each deviation gradient vector. The updated sample set is generated according to the sampling step size by the incremental sparse grid sampling algorithm. The updated sample set and the parameter sample set are concatenated to generate a concatenated sample set, and the polynomial proxy model is incrementally updated based on the concatenated sample set. On the updated polynomial surrogate model, with the interval fields of various parameters as constraints, a line search is performed in the opposite direction of the deviation gradient vector to determine the optimal step size. The parameter values ​​of various parameters are updated according to the optimal step size, and the updated parameter values ​​are substituted into the updated polynomial surrogate model to output the optimized structural response parameters.

[0052] It's important to note that when a deviation occurs between the actual structural response and the surrogate model's prediction, the deviation gradient vector indicates the direction and rate of change of the deviation with respect to the parameters: a larger gradient magnitude indicates a greater influence of that parameter on the deviation, requiring more intensive sampling in that direction. This step first calculates the partial derivatives of the interval deviation with respect to the three types of parameters, obtaining the deviation gradient vector. Then, based on the magnitude of the gradient vector, it adaptively determines the sampling step size for each type of parameter: parameters with large gradient magnitudes use smaller step sizes for fine sampling, while parameters with small gradient magnitudes use larger step sizes for coarse sampling. Finally, an incremental sparse grid sampling algorithm is used to generate an updated sample set in regions with large deviations according to the determined step size, supplementing only a small number of targeted samples without requiring global resampling, significantly improving sampling efficiency.

[0053] Multinomial surrogate models, with their linear parameterization, are well-suited for incremental updates. This step concatenates the updated sample set with the original parameter sample set to generate a new training sample set. Then, based on this concatenated sample set, the coefficients of the surrogate model are incrementally corrected using the least squares method, adjusting only the coefficients related to the newly added samples while retaining most of the original coefficients unchanged. This incremental update method is computationally far less complex than retraining the entire model, enabling model updates to be completed within milliseconds, meeting the real-time requirements of online optimization. The updated surrogate model more accurately reflects the current actual state of the structure.

[0054] The direction of the deviation gradient vector is the direction in which the deviation increases most rapidly; therefore, searching in its opposite direction can minimize the deviation most quickly. This step uses the upper and lower boundaries of the interval fields of various parameters as hard constraints to ensure that parameter adjustments do not exceed the physically possible range. Then, a line search is performed in the opposite direction of the deviation gradient to find the optimal step size that minimizes the deviation. The values ​​of various parameters are updated based on the optimal step size to obtain the optimized parameter combination. Finally, the optimized parameters are substituted into the updated polynomial surrogate model to obtain the optimized structural response parameters. The entire optimization process is completed online, enabling rapid response to changes in the structural state and achieving adaptive adjustment.

[0055] Furthermore, the step of substituting the updated parameter values ​​into the updated polynomial surrogate model to output the optimized structural response parameters includes: Based on the interval field hard constraint boundaries corresponding to each type of parameter, the updated parameter values ​​are subjected to interval folding mapping to obtain a set of mapped parameters, wherein all constraint parameter values ​​are within the range of the interval field after the random field is discretized; The incremental correction coefficients and original intrinsic coefficients of the updated polynomial surrogate model are detected, and a dual-coefficient coupling calculation matrix is ​​constructed accordingly. The mapping parameter set is then input into the dual-coefficient coupling calculation matrix for simultaneous solution processing. Based on the hierarchical matching rules of sparse grid sampling, the simultaneous solution results are subjected to sampling hierarchical calibration processing to obtain the corresponding optimized structural response parameters.

[0056] It's important to note that during gradient optimization, parameter values ​​may exceed the hard constraint boundaries of the interval field, which is physically impossible and could lead to structural safety risks. This step employs an interval folding mapping method to map parameter values ​​exceeding the boundaries: if a parameter value is greater than the upper bound of the interval, it is mapped to the upper bound value; if it is less than the lower bound, it is mapped to the lower bound value. This method ensures that all optimized parameter values ​​remain within the range of the discrete interval field of the random field, guaranteeing the physical rationality of the parameters and the safety of the structure.

[0057] The incrementally updated polynomial surrogate model comprises two parts of coefficients: original intrinsic coefficients, reflecting the global inherent characteristics of the structure; and incremental correction coefficients, reflecting model corrections caused by changes in the local state of the structure. This step integrates these two parts of coefficients to construct a dual-coefficient coupled computation matrix, which can simultaneously consider global characteristics and local corrections. Then, the parameter set after interval folding mapping is input into the dual-coefficient coupled computation matrix for simultaneous solution, yielding preliminary structural response prediction results. This dual-coefficient coupled solution method, compared to using only incremental correction coefficients, better maintains the global consistency of the model and avoids global biases caused by local corrections.

[0058] Incremental sparse grid sampling may be performed at different levels: the initial sample set may use a coarser grid level, while the updated sample set may use a finer grid level. The sampling accuracy varies at different levels, and directly solving the simultaneous equations will introduce errors due to level mismatch. This step performs level calibration on the simultaneous solution results according to the level matching rules of sparse grid sampling, unifying the results from different levels to the same accuracy benchmark. After level calibration, the final optimized structural response parameters have consistent accuracy and reliability, accurately reflecting the true state of the optimized structure and providing a scientific basis for the operation and maintenance adjustments of photovoltaic flexible supports.

[0059] Please see Figure 2 The third embodiment of the present invention provides: An adaptive optimization system for the structural parameters of a photovoltaic flexible support structure, wherein the system comprises: The determination module is used to divide the structural parameters of the photovoltaic flexible support into geometric parameters, material parameters and boundary condition parameters, and for the geometric parameters, the material parameters and the boundary condition parameters, determine the variation range of the photovoltaic flexible support throughout its entire life cycle, and use a random field discretization algorithm to convert the variation range into the corresponding interval field; The generation module is used to generate a corresponding parameter sample set based on the interval field using a sparse grid sampling algorithm, and to embed the upper and lower boundaries of the interval field as hard constraints into the preset basis functions of the polynomial. Combined with the parameter sample set, a polynomial surrogate model of the interval field constraint is constructed accordingly. The calculation module is used to calculate the interval sensitivity corresponding to the geometric parameters, the material parameters and the boundary condition parameters based on the polynomial surrogate model, and to detect the corresponding actual structural response parameters during the actual operation of the photovoltaic flexible support. The optimization module is used to compare the actual structural response parameters with the interval sensitivity to calculate the corresponding interval deviation. When the interval deviation exceeds a preset deviation threshold, an incremental sparse grid sampling algorithm is used to generate a corresponding updated sample set based on the interval deviation and the parameter sample set. The updated sample set is then used to perform online optimization processing on the actual structural response parameters to generate corresponding optimized structural response parameters.

[0060] Furthermore, the determining module is specifically used for: Based on the historical operating data of the photovoltaic flexible support, a Markov transition matrix of the structural parameters changing over time is constructed, and combined with a preset spatial covariance function, the spatiotemporal joint prior distribution of the structural parameters is constructed. The structure of the photovoltaic flexible support is divided into several discrete units by the random field discretization algorithm, and the parameter values ​​of each discrete unit are used as random variables of the conditional random field. The spatiotemporal joint prior distribution is used as the potential function of the conditional random field, and the belief propagation algorithm is used to solve the conditional random field to obtain the marginal probability distribution of the parameter values ​​of each discrete unit. The confidence intervals of each discrete unit are calculated based on the marginal probability distribution, and the confidence intervals are spatially spliced ​​together to generate the interval field.

[0061] Furthermore, the determining module is specifically used for: Each discrete unit is used as a graph node, the physical connection relationship between each discrete unit is used as a graph edge, and the weight of the corresponding graph edge is determined according to the coupling coefficient between adjacent discrete units to construct the corresponding uncertainty propagation graph. Each confidence interval is assigned to the interior of each node in the uncertainty propagation graph, and the attention coefficient of each node to all its neighboring nodes is calculated using a graph attention network. Based on the attention coefficient and the edge weight of the node, the confidence interval of each node is weighted and propagated to its neighboring nodes to generate the corresponding spatial interval. Based on the Markov transition matrix, the actual parameter intervals of each discrete unit are calculated, and the spatial intervals and the actual parameter intervals are weighted and fused to generate the interval field.

[0062] Furthermore, the generation module is specifically used for: The interval field of each discrete unit is decomposed into a central field and a radius field, and the spatial distribution characteristics of the central field and the radius field are detected respectively. The central field is the arithmetic mean of the upper and lower bounds of the interval field, and the radius field is half of the difference between the upper and lower bounds of the interval field. Based on the spatial distribution characteristics, a central field basis function is constructed according to the central field, and a radius field basis function is constructed according to the radius field. The central field basis function and the radius field basis function are then combined by Cartesian product to obtain a set of polynomial basis functions with center-radius separation. Using the parameter sample set as training data, the prediction residuals of the central field and the radius field are calculated respectively. With all output values ​​of the radius field being greater than or equal to zero as hard constraints, the residuals are input into the polynomial basis function set to construct the polynomial surrogate model.

[0063] Furthermore, the calculation module is specifically used for: The first-order partial derivatives of the geometric parameters, material parameters, and boundary condition parameters are solved using the polynomial proxy model to obtain the partial derivative polynomial operators corresponding to each parameter type. The parameter space correlation matrix between each discrete unit is constructed based on the preset spatial covariance function, and the partial derivative polynomial operator is multiplied with the parameter space correlation matrix to generate the corresponding extended partial derivative operator. The range of values ​​of the extended partial derivative operator in its corresponding parameter interval is calculated using interval arithmetic method to obtain the upper and lower bounds of the interval sensitivity of each discrete unit. The upper and lower bounds of the interval sensitivity are then integrated to generate the corresponding interval sensitivity.

[0064] Furthermore, the optimization module is specifically used for: The partial derivatives of the interval deviation with respect to the geometric parameters, the material parameters, and the boundary condition parameters are calculated to obtain the deviation gradient vector. The sampling step size of each type of parameter is determined according to the magnitude of each deviation gradient vector. The updated sample set is generated according to the sampling step size by the incremental sparse grid sampling algorithm. The updated sample set and the parameter sample set are concatenated to generate a concatenated sample set, and the polynomial proxy model is incrementally updated based on the concatenated sample set. On the updated polynomial surrogate model, with the interval fields of various parameters as constraints, a line search is performed in the opposite direction of the deviation gradient vector to determine the optimal step size. The parameter values ​​of various parameters are updated according to the optimal step size, and the updated parameter values ​​are substituted into the updated polynomial surrogate model to output the optimized structural response parameters.

[0065] Furthermore, the optimization module is specifically used for: Based on the interval field hard constraint boundaries corresponding to each type of parameter, the updated parameter values ​​are subjected to interval folding mapping to obtain a set of mapped parameters, wherein all constraint parameter values ​​are within the range of the interval field after the random field is discretized; The incremental correction coefficients and original intrinsic coefficients of the updated polynomial surrogate model are detected, and a dual-coefficient coupling calculation matrix is ​​constructed accordingly. The mapping parameter set is then input into the dual-coefficient coupling calculation matrix for simultaneous solution processing. Based on the hierarchical matching rules of sparse grid sampling, the simultaneous solution results are subjected to sampling hierarchical calibration processing to obtain the corresponding optimized structural response parameters.

[0066] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the adaptive optimization method for photovoltaic flexible support structure parameters as described above.

[0067] The fifth embodiment of the present invention provides a readable storage medium storing a computer program thereon, wherein the program, when executed by a processor, implements the adaptive optimization method for photovoltaic flexible support structure parameters as described above.

[0068] In summary, the adaptive optimization method and system for photovoltaic flexible support structure parameters provided in the above embodiments of the present invention can update the sample set and complete online optimization by means of incremental sparse grid sampling when the deviation exceeds the threshold. This effectively avoids the problem of deviation between control actions and actual needs, significantly improves the optimization efficiency of photovoltaic flexible support structure parameters, and ensures the accuracy and stability of support operation and control under extreme working conditions.

[0069] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0070] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0071] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0072] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0073] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0074] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. An adaptive optimization method for the structural parameters of a photovoltaic flexible support structure, characterized in that, The method includes: The structural parameters of the photovoltaic flexible support are divided into geometric parameters, material parameters, and boundary condition parameters. For the geometric parameters, material parameters, and boundary condition parameters, the variation range of the photovoltaic flexible support throughout its entire life cycle is determined. The variation range is then converted into a corresponding interval field using a random field discretization algorithm. Based on the interval field, a sparse grid sampling algorithm is used to generate a corresponding parameter sample set, and the upper and lower boundaries of the interval field are embedded as hard constraints into the preset basis functions of the polynomial. Combined with the parameter sample set, a polynomial surrogate model with interval field constraints is constructed. Based on the polynomial surrogate model, the interval sensitivity corresponding to the geometric parameters, the material parameters and the boundary condition parameters is calculated respectively, and the corresponding actual structural response parameters are detected during the actual operation of the photovoltaic flexible support. The actual structural response parameters are compared with the interval sensitivity to calculate the corresponding interval deviation. When the interval deviation exceeds a preset deviation threshold, an incremental sparse grid sampling algorithm is used to generate a corresponding updated sample set based on the interval deviation and the parameter sample set. The actual structural response parameters are then optimized online using the updated sample set to generate corresponding optimized structural response parameters. The step of converting the range of change into a corresponding interval field using a random field discretization algorithm includes: Based on the historical operating data of the photovoltaic flexible support, a Markov transition matrix of the structural parameters changing over time is constructed, and combined with a preset spatial covariance function, the spatiotemporal joint prior distribution of the structural parameters is constructed. The structure of the photovoltaic flexible support is divided into several discrete units by the random field discretization algorithm, and the parameter values ​​of each discrete unit are used as random variables of the conditional random field. The spatiotemporal joint prior distribution is used as the potential function of the conditional random field, and the belief propagation algorithm is used to solve the conditional random field to obtain the marginal probability distribution of the parameter values ​​of each discrete unit. The confidence intervals of each discrete unit are calculated according to the marginal probability distribution, and the confidence intervals are spatially spliced ​​to generate the interval field. The step of calculating the interval sensitivity corresponding to the geometric parameters, material parameters, and boundary condition parameters based on the polynomial surrogate model includes: The first-order partial derivatives of the geometric parameters, material parameters, and boundary condition parameters are solved using the polynomial proxy model to obtain the partial derivative polynomial operators corresponding to each parameter type. The parameter space correlation matrix between each discrete unit is constructed based on the preset spatial covariance function, and the partial derivative polynomial operator is multiplied with the parameter space correlation matrix to generate the corresponding extended partial derivative operator. The range of values ​​of the extended partial derivative operator in its corresponding parameter interval is calculated using interval arithmetic method to obtain the upper and lower bounds of the interval sensitivity of each discrete unit. The upper and lower bounds of the interval sensitivity are then integrated to generate the corresponding interval sensitivity.

2. The adaptive optimization method for photovoltaic flexible support structure parameters according to claim 1, characterized in that, The step of spatially stitching together the confidence intervals to generate the corresponding interval field includes: Each discrete unit is used as a graph node, the physical connection relationship between each discrete unit is used as a graph edge, and the weight of the corresponding graph edge is determined according to the coupling coefficient between adjacent discrete units to construct the corresponding uncertainty propagation graph. Each confidence interval is assigned to the interior of each node in the uncertainty propagation graph, and the attention coefficient of each node to all its neighboring nodes is calculated using a graph attention network. Based on the attention coefficient and the edge weight of the node, the confidence interval of each node is weighted and propagated to its neighboring nodes to generate the corresponding spatial interval. Based on the Markov transition matrix, the actual parameter intervals of each discrete unit are calculated, and the spatial intervals and the actual parameter intervals are weighted and fused to generate the interval field.

3. The adaptive optimization method for photovoltaic flexible support structure parameters according to claim 1, characterized in that, The step of embedding the upper and lower boundaries of the interval field as hard constraints into the preset basis functions of the polynomial, and constructing a polynomial surrogate model of the interval field constraints in conjunction with the parameter sample set, includes: The interval field of each discrete unit is decomposed into a central field and a radius field, and the spatial distribution characteristics of the central field and the radius field are detected respectively. The central field is the arithmetic mean of the upper and lower bounds of the interval field, and the radius field is half of the difference between the upper and lower bounds of the interval field. Based on the spatial distribution characteristics, a central field basis function is constructed according to the central field, and a radius field basis function is constructed according to the radius field. The central field basis function and the radius field basis function are then combined by Cartesian product to obtain a set of polynomial basis functions with center-radius separation. Using the parameter sample set as training data, the prediction residuals of the central field and the radius field are calculated respectively. With all output values ​​of the radius field being greater than or equal to zero as hard constraints, the residuals are input into the polynomial basis function set to construct the polynomial surrogate model.

4. The adaptive optimization method for photovoltaic flexible support structure parameters according to claim 1, characterized in that, The step of performing online optimization of the actual structural response parameters using the updated sample set to generate corresponding optimized structural response parameters includes: The partial derivatives of the interval deviation with respect to the geometric parameters, the material parameters, and the boundary condition parameters are calculated to obtain the deviation gradient vector. The sampling step size of each type of parameter is determined according to the magnitude of each deviation gradient vector. The updated sample set is generated according to the sampling step size by the incremental sparse grid sampling algorithm. The updated sample set and the parameter sample set are concatenated to generate a concatenated sample set, and the polynomial proxy model is incrementally updated based on the concatenated sample set. On the updated polynomial surrogate model, with the interval fields of various parameters as constraints, a line search is performed in the opposite direction of the deviation gradient vector to determine the optimal step size. The parameter values ​​of various parameters are updated according to the optimal step size, and the updated parameter values ​​are substituted into the updated polynomial surrogate model to output the optimized structural response parameters.

5. The adaptive optimization method for photovoltaic flexible support structure parameters according to claim 4, characterized in that, The step of substituting the updated parameter values ​​into the updated polynomial surrogate model to output the optimized structural response parameters includes: Based on the interval field hard constraint boundaries corresponding to each type of parameter, the updated parameter values ​​are subjected to interval folding mapping to obtain a set of mapped parameters, wherein all constraint parameter values ​​are within the range of the interval field after the random field is discretized; The incremental correction coefficients and original intrinsic coefficients of the updated polynomial surrogate model are detected, and a dual-coefficient coupling calculation matrix is ​​constructed accordingly. The mapping parameter set is then input into the dual-coefficient coupling calculation matrix for simultaneous solution processing. Based on the hierarchical matching rules of sparse grid sampling, the simultaneous solution results are subjected to sampling hierarchical calibration processing to obtain the corresponding optimized structural response parameters.

6. A photovoltaic flexible support structure parameter adaptive optimization system, characterized in that, The system is used to implement the adaptive optimization method for photovoltaic flexible support structure parameters as described in any one of claims 1 to 5, the system comprising: The determination module is used to divide the structural parameters of the photovoltaic flexible support into geometric parameters, material parameters and boundary condition parameters, and for the geometric parameters, the material parameters and the boundary condition parameters, determine the variation range of the photovoltaic flexible support throughout its entire life cycle, and use a random field discretization algorithm to convert the variation range into the corresponding interval field; The generation module is used to generate a corresponding parameter sample set based on the interval field using a sparse grid sampling algorithm, and to embed the upper and lower boundaries of the interval field as hard constraints into the preset basis functions of the polynomial. Combined with the parameter sample set, a polynomial surrogate model of the interval field constraint is constructed accordingly. The calculation module is used to calculate the interval sensitivity corresponding to the geometric parameters, the material parameters and the boundary condition parameters based on the polynomial surrogate model, and to detect the corresponding actual structural response parameters during the actual operation of the photovoltaic flexible support. The optimization module is used to compare the actual structural response parameters with the interval sensitivity to calculate the corresponding interval deviation. When the interval deviation exceeds a preset deviation threshold, an incremental sparse grid sampling algorithm is used to generate a corresponding updated sample set based on the interval deviation and the parameter sample set. The updated sample set is then used to perform online optimization processing on the actual structural response parameters to generate corresponding optimized structural response parameters.

7. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the adaptive optimization method for photovoltaic flexible support structure parameters as described in any one of claims 1 to 5.

8. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the adaptive optimization method for photovoltaic flexible support structure parameters as described in any one of claims 1 to 5.

Images