Flywheel energy storage system variable thickness flywheel molded line optimization method, system, equipment and medium

Abstract

The invention relates to the technical field of flywheel energy storage system structure optimization, and discloses a flywheel energy storage system variable thickness flywheel profile optimization method, system, device and medium, and the method comprises the steps: building a dual-target optimization model, and taking the maximum unit mass energy storage density and the minimum maximum centrifugal stress as targets; the geometric boundary of the flywheel, the orthotropic property of the composite material and the lamination process constraint are considered, the reasonable and feasible design is ensured, and the failure states such as interlayer stripping or fiber fracture are reduced. And further parameterizing flywheel thickness distribution by adopting a third-order B-spline function, forming a design parameter space in combination with a fiber laying angle, setting a non-uniform node vector and an adjacent coefficient difference upper limit constraint, and ensuring radial smooth change of the thickness. An optimized data set is generated through finite element simulation based on a parameterized model, and a foundation is laid for physical information neural network training. The method provides a scientific, efficient and reliable design optimization scheme for the flywheel energy storage system, and assists energy transformation and sustainable development.

Description

Technical Field

[0001] This invention relates to the field of flywheel energy storage system structure optimization technology, and in particular to a method, system, equipment and medium for optimizing the profile of a flywheel energy storage system with variable thickness. Background Technology

[0002] Currently, flywheel energy storage systems are considered a relatively efficient and clean physical energy storage method. These systems use a high-speed rotating flywheel to convert electrical energy into kinetic energy for storage. Flywheel energy storage systems offer many advantages, such as high power density, short charge / discharge response time, long cycle life, and no chemical pollution. Therefore, due to these advantages, flywheel energy storage systems have been widely used in areas such as grid frequency regulation, regenerative braking in new energy vehicles, industrial uninterruptible power supplies (UPS), and aerospace emergency power supply.

[0003] With the accelerating global energy transition, the demand for energy storage systems with high energy density and high security is increasing. As a result, the use of flywheel energy storage systems is also increasing. Compared with traditional metal materials, composite materials have the characteristics of high specific strength, high specific modulus and anisotropic designability, making them the preferred material for flywheels in high-performance flywheel energy storage systems.

[0004] Therefore, the current application of flywheel energy storage systems has shifted to the structural design of composite material flywheels. In the design of composite materials, variable thickness flywheel profiles are widely used. The key to variable thickness flywheels is to simultaneously optimize the energy storage density per unit mass and centrifugal stress to meet the requirements of high energy storage and safe operation.

[0005] However, current optimization methods for variable thickness profiles still face many challenges. For example, many existing models are based on simplified models of metallic materials, failing to fully consider the orthotropic anisotropy of composite materials and the influence of fiber layup angle on mechanical properties. This leads to inaccurate stress analysis, resulting in failures such as interlaminar delamination or fiber breakage. Furthermore, traditional methods often employ single-objective optimization or simple weighted summation to balance two objectives. Such designs cannot meet the requirements of high rotational speeds and radial stress constraints in composite materials, making true synergistic optimization difficult to achieve. Summary of the Invention

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] Therefore, this invention provides a method, system, equipment, and medium for optimizing the profile of a flywheel energy storage system with variable thickness, which can solve the problems in the prior art, such as inaccurate stress analysis and difficulty in achieving dual-objective collaborative optimization, due to insufficient consideration of the characteristics of composite materials and the influence of fiber layup angle.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for optimizing the profile of a flywheel energy storage system with variable thickness, comprising: A dual-objective optimization model is established with the objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress. Construct a constraint system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; The thickness distribution of the flywheel along the radial direction is parameterized using a third-order B-spline function. The B-spline coefficients are used as optimization variables, and a design parameter space is formed by combining the fiber layup angle to establish a parameterized model. Based on the parametric model, a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress is generated through finite element simulation. Construct a physical information neural network, and train the physical information neural network using the dataset until its prediction results meet the preset stress and displacement error limits; A proximal policy optimization algorithm is constructed. During the policy iteration process of the proximal policy optimization algorithm, the trained physical information neural network is called to provide stress and energy storage feedback in real time until the policy converges and the optimal B-spline coefficient and the optimal fiber layup angle are output.

[0009] As a preferred embodiment of the variable thickness flywheel profile optimization method for the flywheel energy storage system described in this invention, the constraint system includes: Set the minimum and maximum values ​​of the flywheel's inner diameter, outer diameter, and thickness function; Set the elastic modulus, principal Poisson's ratio, secondary Poisson's ratio, shear modulus, and density of the composite material in the principal and secondary directions, as well as the allowable stresses for tensile stress in the principal direction, tensile stress in the secondary direction, and interlaminar shear. Based on orthotropic elasticity, the equilibrium equations of radial and circumferential forces, strain-displacement geometric relations, and stress-strain constitutive relations are established. A continuity constraint is set on the fiber layup angle to limit drastic changes in the layup angle caused by abrupt changes in thickness in adjacent areas.

[0010] This preferred solution ensures the rationality and feasibility of the flywheel design, enabling more scientific and accurate design and calculation. This allows the optimized flywheel to meet various mechanical performance requirements while maximizing energy storage efficiency, providing strong technical support for the practical application of flywheel energy storage systems.

[0011] As a preferred embodiment of the variable thickness flywheel profile optimization method for the flywheel energy storage system described in this invention, the step of parameterizing the flywheel thickness distribution using a third-order B-spline function includes: Based on the characteristics of a large radial stress gradient near the inner diameter of the flywheel and a significant circumferential stress near the outer edge, a non-uniformly distributed node vector is set. The thickness function is represented as a linear combination of B-spline basis functions and coefficients, where the coefficients are treated as continuous optimization variables. An upper limit constraint on the difference between adjacent B-spline coefficients is applied to ensure that the thickness varies smoothly in the radial direction.

[0012] As a preferred embodiment of the variable thickness flywheel profile optimization method for the flywheel energy storage system described in this invention, the dataset includes: An improved Latin hypercube sampling method is used to extract sample points in the design space composed of B-spline coefficients. Increase the sample density in the region near the inner diameter of the flywheel; For each sample point, a finite element model of the composite laminate structure is established, the orthogonal anisotropic material properties and laminate section are defined, a rated angular velocity load is applied, and the stress field and displacement field are obtained by solving. Multiple discrete locations are selected on the radial coordinate, and the radial stress, circumferential stress, radial displacement, and interlaminar shear stress at each location are extracted to form training samples.

[0013] As a preferred embodiment of the variable thickness flywheel profile optimization method for the flywheel energy storage system described in this invention, the training loss of the physical information neural network includes four terms: The first item is data loss, which measures the deviation between the network output and the finite element simulation results in each stress component and displacement. The second term is physical loss, which measures whether the network output satisfies the equilibrium equation and constitutive relation of orthotropic materials. The third term is boundary loss, which is used to force the satisfaction of the constraint conditions that the displacement at the inner diameter is zero, the stress at the outer diameter is free, and the interlayer shear stress is at the boundary. The fourth item is the shear penalty loss, which is used to impose an additional penalty when the interlaminar shear stress exceeds the allowable value; The physical information neural network takes radial coordinates, B-spline coefficients and fiber layup angle as inputs, and radial stress, circumferential stress, radial displacement and interlaminar shear stress as outputs. During the training process, orthogonal anisotropic elastic mechanical control equations, boundary conditions and interlaminar shear strength constraints are introduced as physical loss terms.

[0014] As a preferred embodiment of the variable thickness flywheel profile optimization method for the flywheel energy storage system described in this invention, the near-end policy optimization algorithm includes a policy network and a value network: The state space of the near-end strategy optimization algorithm consists of normalized B-spline coefficients and normalized fiber layup angles. The action space includes the adjustment amount of the B-spline coefficients and the fine-tuning amount of the fiber layup angles. The reward function is constructed based on the degree of improvement of the energy storage density per unit mass and whether the stress in each direction exceeds the corresponding allowable value. The policy network receives the current state, processes it through multiple fully connected layers, and outputs the mean and standard deviation of the actions to form a probability distribution of consecutive actions. The value network contains two parallel processing channels. The first channel receives the thickness parameters and evaluates their corresponding energy storage potential, while the second channel receives the thickness parameters and fiber layup angle and evaluates their corresponding stress safety. During the strategy update process, if the interlaminar shear stress predicted by the physical information neural network exceeds the allowable value, the action sampling logic is adjusted to prioritize actions that reduce the risk of interlaminar shear.

[0015] As a preferred embodiment of the variable thickness flywheel profile optimization method for the flywheel energy storage system described in this invention, it further includes a verification step: The optimal B-spline coefficients and optimal fiber layup angle obtained after convergence are inversely normalized to obtain the actual thickness distribution function and fiber layup angle. A composite laminate flywheel model was established in finite element software. The mesh was generated using an element type that adapted to the interlayer interface. The thickness distribution and the angle of the alternately laid fibers were set according to the inverse normalization results. Apply a rated angular velocity load and calculate the radial stress, circumferential stress, interlayer shear stress, and energy density per unit mass of the flywheel. The finite element calculation results are compared with the predicted values ​​of the physical information neural network. If all errors are within the preset tolerance range, the optimization results are confirmed to be effective.

[0016] Secondly, the present invention provides a variable thickness flywheel profile optimization system for flywheel energy storage systems, comprising: The model building module is used to build a bi-objective optimization model with the optimization objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress. The constraint establishment module is used to construct a constraint condition system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; The parameterization module is used to parameterize the thickness distribution of the flywheel along the radial direction using a third-order B-spline function. The B-spline coefficients are used as optimization variables, and combined with the fiber layup angle to form a design parameter space and establish a parameterized model. The dataset acquisition module is used to generate a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress based on the parametric model and through finite element simulation. The neural network construction module is used to construct a physical information neural network and train the physical information neural network using the dataset until its prediction results meet the preset stress and displacement error limits. The optimization module is used to construct the near-end policy optimization algorithm. During the policy iteration process of the near-end policy optimization algorithm, the trained physical information neural network is called to provide stress and energy storage feedback in real time until the policy converges and outputs the optimal B-spline coefficient and the optimal fiber layup angle.

[0017] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0018] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0019] Compared with existing technologies, the beneficial effects of this invention are that it proposes a variable-thickness flywheel profile optimization method for flywheel energy storage systems. By establishing a dual-objective optimization model, with the objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress, the method considers flywheel geometric boundaries, orthogonal anisotropy of composite materials, and lamination process constraints to ensure reasonable and feasible design and reduce failure states such as interlaminar delamination or fiber breakage. Furthermore, a third-order B-spline function is used to parameterize the flywheel thickness distribution, combined with the fiber layup angle to form a design parameter space. Non-uniform node vectors and upper limits for adjacent coefficient differences are set to ensure smooth radial thickness variation. Based on the parameterized model, an optimization dataset is generated through finite element simulation, laying the foundation for training a physical information neural network. This invention provides a scientific, efficient, and reliable design optimization scheme for flywheel energy storage systems, contributing to energy transition and sustainable development. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a method for optimizing the profile of a flywheel energy storage system with variable thickness, as provided in one embodiment of the present invention.

[0022] Figure 2 This is an internal structure diagram of an electronic device for a flywheel energy storage system with a variable thickness flywheel profile optimization method provided in one embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for optimizing the profile of a flywheel energy storage system with variable thickness, comprising: This invention provides a method that can effectively solve the problems mentioned above. The following will describe in detail how to implement the variable thickness flywheel profile optimization method for the flywheel energy storage system with multiple embodiments. Figure 1 A flowchart illustrating a method for optimizing the profile of a flywheel energy storage system with variable thickness is shown, including: S101, establish a dual-objective optimization model with the optimization objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress; It should be noted that establishing a dual-objective optimization model with the optimization objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress means that in the design process of flywheel energy storage system, the two core indicators of energy storage efficiency and structural safety performance are considered simultaneously, and the synergistic optimization between the two is achieved through scientific modeling.

[0025] Among them, energy storage density per unit mass refers to the kinetic energy that the flywheel can store per unit mass at its rated speed. The higher the value, the more energy the flywheel can store under the same weight. This is of great significance for improving the overall energy efficiency of the system, reducing equipment weight, and reducing manufacturing costs. Maximum centrifugal stress refers to the maximum stress value generated inside the material due to centrifugal force when the flywheel is rotating at high speed. If this stress exceeds the allowable strength of the material, it will lead to failure modes such as fiber breakage, matrix cracking, or interlaminar delamination, which seriously threatens operational safety.

[0026] In one alternative implementation, flywheel energy storage systems are widely used in practical engineering applications, such as grid frequency regulation, rail transit braking energy recovery, industrial uninterruptible power supplies, and aerospace emergency power supply, where extremely high requirements are placed on response speed and cycle life. These applications not only require flywheels to have high energy density but also impose stringent standards on their structural reliability under long-term high-speed operation. Traditional design methods often use uniform thickness or empirical profiles, making it difficult to balance energy storage performance with stress distribution. Especially when using composite materials, due to their orthotropic characteristics, the mechanical response varies significantly in different directions. If optimization is performed with only a single objective, it is very easy to cause stress exceeding the limit in a certain direction, leading to early failure.

[0027] The dual-objective optimization model constructed in this invention is a systematic design to address the aforementioned problems. The model first explicitly sets energy storage density per unit mass as the primary optimization objective. By adjusting the radial thickness distribution of the flywheel, the ratio of rotational inertia to total mass is increased, thereby enhancing energy storage capacity without increasing overall weight. Simultaneously, maximum centrifugal stress is used as a key constraint objective, focusing on monitoring whether radial stress, circumferential stress, and interlaminar shear stress exceed the allowable limits of the composite material in the corresponding directions. There is an inherent contradiction between these two objectives: increasing thickness can improve rotational inertia but exacerbates stress concentration caused by centrifugal loads; thinning the structure helps reduce stress but weakens energy storage capacity. Therefore, an optimal balance must be sought between the two through a multi-objective collaborative mechanism.

[0028] In an alternative implementation, the bi-objective optimization model is not simply a weighted sum of the two objectives, but rather embedded within a reinforcement learning-based intelligent optimization framework. The model dynamically evaluates the energy storage potential and stress risk under the current design state using a proximal policy optimization algorithm, and guides the search direction based on a preset reward function: providing positive incentives when the energy density per unit mass increases, and imposing penalties when any stress component approaches or exceeds the allowable value, especially setting an additional penalty term for interlaminar shear stress to emphasize the protection of weak interfaces in the composite material. This mechanism ensures that the optimization process not only approaches the Pareto optimal frontier but also guarantees that the obtained solution meets the safety margin requirements of actual engineering projects.

[0029] Furthermore, this model fully integrates the physical properties of composite materials. It treats the fiber layup angle as an independent design variable, combining it with the thickness distribution parameterized by a third-order B-spline function to form a high-dimensional design space, allowing for coordinated adjustment of the thickness profile and layup direction. For example, high-strength fibers can be placed in circumferential high-stress regions to bear the main tensile stress, while radial stress concentration can be alleviated near the inner diameter through local thickening or optimization of the thickness gradient, thus achieving a deep integration of material properties and structural morphology. This integrated design approach effectively avoids stress misjudgment and structural redundancy caused by neglecting anisotropy in traditional methods.

[0030] It should be noted that this dual-objective optimization model achieves a precise trade-off between high energy density and high structural safety in flywheel energy storage systems, providing a theoretical basis and technical path for the high-performance design of composite material flywheels, and significantly improving the comprehensive performance and engineering applicability of flywheel energy storage systems under complex working conditions.

[0031] S102, Construct a constraint system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; In embodiments of the present invention, the constraint system includes: Set the minimum and maximum values ​​of the flywheel's inner diameter, outer diameter, and thickness function; Set the elastic modulus, principal Poisson's ratio, secondary Poisson's ratio, shear modulus, and density of the composite material in the principal and secondary directions, as well as the allowable stresses for tensile stress in the principal direction, tensile stress in the secondary direction, and interlaminar shear. Based on orthotropic elasticity, the equilibrium equations of radial and circumferential forces, strain-displacement geometric relations, and stress-strain constitutive relations are established. A continuity constraint is set on the fiber layup angle to limit drastic changes in the layup angle caused by abrupt changes in thickness in adjacent areas.

[0032] Specifically, in some embodiments, the inner diameter of the flywheel can be set to the size of the fixed mounting hole, the outer diameter to the maximum allowable envelope size for high-speed rotation, and the minimum value of the thickness function can be specified to be no less than the thickness of a single layer of prepreg in the composite material to ensure process feasibility, and the maximum value can be specified to be no more than the structural space limit to avoid interference with the shell.

[0033] In some embodiments, the elastic modulus of the composite material in the main fiber direction can be set to the typical value of high-modulus carbon fiber, the elastic modulus in the secondary direction can be determined based on the properties of the resin matrix, the main Poisson's ratio reflects the transverse shrinkage characteristics in the fiber direction, the secondary Poisson's ratio reflects the coupled deformation in the vertical direction, the shear modulus can be calibrated by interlaminar shear test, the density can be the measured average value, the allowable tensile stress in the main direction can be taken as the safe reduction value of the fiber breaking strength, the allowable tensile stress in the secondary direction can be set based on the matrix cracking threshold, and the allowable interlaminar shear stress can be determined based on the short beam shear test results.

[0034] In some embodiments, equilibrium differential equations containing radial and circumferential stress coupling terms can be established based on orthotropic elasticity. The strain-displacement relationship adopts axisymmetric small deformation geometric equations, and the stress-strain constitutive relationship adopts orthotropic generalized Hooke's law and explicitly expresses the stiffness matrix in each direction.

[0035] In some embodiments, an upper limit constraint can be set on the rate of change of angle between adjacent radial segments of the fiber layup angle, for example, limiting the layup angle change to no more than 5 degrees per millimeter of radial distance, to prevent layup wrinkles or fiber buckling caused by sudden changes in thickness.

[0036] Among them, the thickness function is made continuously differentiable through third-order B-spline parameterization, the fiber lay-up angle is used as an independent design variable and together with the B-spline coefficients constitute the optimization space, and all material parameters and geometric boundaries are directly used for the embedding of control equations in finite element modeling and physical information neural networks.

[0037] It should be noted that the advantage of S102 is that it ensures that the optimization results meet the mechanical properties of composite materials and the requirements of actual manufacturing processes, preventing unrealizable or easily failed structural forms.

[0038] S103, a third-order B-spline function is used to parameterize the thickness distribution of the flywheel along the radial direction, the B-spline coefficient is used as the optimization variable, and the design parameter space is formed by combining the fiber layup angle to establish a parameterized model; In this embodiment of the invention, a third-order B-spline function is used to parameterize the flywheel thickness distribution, including: Based on the characteristics of a large radial stress gradient near the inner diameter of the flywheel and a significant circumferential stress near the outer edge, a non-uniformly distributed node vector is set. Here, the node vector is set to be a non-equidistant distribution with denser nodes near the inner diameter to improve the accuracy of describing the thickness variation in this region. The thickness function is represented as a linear combination of B-spline basis functions and coefficients, where the coefficients are treated as continuous optimization variables. An upper limit constraint on the difference between adjacent B-spline coefficients is applied to ensure that the thickness varies smoothly in the radial direction.

[0039] Specifically, in some embodiments, based on the characteristics of the flywheel being subjected to significant radial stress gradient in the inner diameter region and high circumferential tensile stress in the outer edge region under high-speed rotation, the node vector of the third-order B-spline can be designed as a non-uniform distribution in the parametric modeling. For example, the nodes can be densely arranged in the range from the inner diameter to 30% of the radius to improve the representation ability of local thickness changes, while the node spacing in the outer edge region can be gradually sparsed to match the characteristics of gradual stress change.

[0040] In some embodiments, the flywheel thickness function can be expressed as a linear combination of a third-order B-spline basis function and the corresponding control coefficients, where each control coefficient participates in the multi-objective optimization process as a continuously adjustable design variable, directly determining the structural thickness at that radial position.

[0041] In some embodiments, a hard constraint can be imposed on adjacent control coefficients with an absolute difference not exceeding 0.2 mm. This value is determined by the maximum interlayer step height allowed by the automated fiber placement process of composite materials, thereby forcing a monotonous and smooth transition of thickness in the radial direction and avoiding fiber buckling or resin enrichment defects caused by abrupt changes.

[0042] Among them, the third-order B-spline function has C² continuity, which ensures the continuity of the first and second derivatives of the thickness function, satisfying the strict requirements of geometric smoothness for the dynamic analysis of the rotating body structure.

[0043] It should be noted that the advantage of S103 is that it can accurately describe complex variable thickness profiles with a small number of continuous variables, while being compatible with fiber angle design, reducing the optimization dimensionality and improving search efficiency.

[0044] S104, based on a parametric model, generates a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress through finite element simulation; In this embodiment of the invention, the dataset includes: An improved Latin hypercube sampling method is used to extract sample points in the design space composed of B-spline coefficients. Increase the sample density in the region near the inner diameter of the flywheel. This means that the radial stress is more sensitive to changes in thickness, so the sampling is made more dense in the inner diameter region. For each sample point, a finite element model of the composite laminate structure is established, the orthogonal anisotropic material properties and laminate section are defined, a rated angular velocity load is applied, and the stress field and displacement field are obtained by solving. Multiple discrete locations are selected on the radial coordinate, and the radial stress, circumferential stress, radial displacement, and interlaminar shear stress at each location are extracted to form training samples.

[0045] It should be noted that the advantage of S104 is that it obtains high-fidelity training samples that cover key failure modes, providing physically accurate supervision signals for subsequent data-driven models.

[0046] S105, Construct a physical information neural network, and train the physical information neural network using the dataset until its prediction results meet the preset stress and displacement error limits. In this embodiment of the invention, the training loss of the physical information neural network includes four items: The first item is data loss, which measures the deviation between the network output and the finite element simulation results in each stress component and displacement. The second term is physical loss, which measures whether the network output satisfies the equilibrium equation and constitutive relation of orthotropic materials. The third term is boundary loss, which is used to force the satisfaction of the constraint conditions that the displacement at the inner diameter is zero, the stress at the outer diameter is free, and the interlayer shear stress is at the boundary. The fourth item is the shear penalty loss, which is used to impose an additional penalty when the interlaminar shear stress exceeds the allowable value; In this embodiment of the invention, the physical information neural network takes radial coordinates, B-spline coefficients and fiber layup angle as inputs, and radial stress, circumferential stress, radial displacement and interlaminar shear stress as outputs. During the training process, orthogonal anisotropic elastic mechanical control equations, boundary conditions and interlaminar shear strength constraints are introduced as physical loss terms.

[0047] It should be noted that the advantage of S105 is that it provides a proxy model that can quickly and accurately predict multiple stress fields and displacements, replacing high-cost finite element calculations and supporting efficient optimization iterations.

[0048] S106. Construct a near-end strategy optimization algorithm. During the strategy iteration process of the near-end strategy optimization algorithm, call the trained physical information neural network to provide stress and energy storage feedback in real time until the strategy converges and outputs the optimal B-spline coefficient and the optimal fiber layup angle.

[0049] In this embodiment of the invention, the near-end policy optimization algorithm includes a policy network and a value network: The state space of the near-end policy optimization algorithm consists of normalized B-spline coefficients and normalized fiber layup angles. The action space includes the adjustment amount of the B-spline coefficients and the fine-tuning amount of the fiber layup angles. The reward function is constructed based on the degree of improvement of energy storage density per unit mass and whether the stress in each direction exceeds the corresponding allowable value. The policy network receives the current state, processes it through multiple fully connected layers, and outputs the mean and standard deviation of the actions to form a probability distribution of consecutive actions. The value network contains two parallel processing channels. The first channel receives the thickness parameters and evaluates their corresponding energy storage potential, while the second channel receives the thickness parameters and fiber layup angle and evaluates their corresponding stress safety. During the strategy update process, if the interlaminar shear stress predicted by the physical information neural network exceeds the allowable value, the action sampling logic is adjusted to prioritize actions that reduce the risk of interlaminar shear.

[0050] In this embodiment of the invention, a verification step is also included: The optimal B-spline coefficients and optimal fiber layup angle obtained after convergence are inversely normalized to obtain the actual thickness distribution function and fiber layup angle. A composite laminate flywheel model was established in finite element software. The mesh was generated using an element type that adapted to the interlayer interface. The thickness distribution and the angle of the alternately laid fibers were set according to the inverse normalization results. Apply a rated angular velocity load and calculate the radial stress, circumferential stress, interlayer shear stress, and energy density per unit mass of the flywheel. The finite element calculation results are compared with the predicted values ​​of the physical information neural network. If all errors are within the preset tolerance range, the optimization results are confirmed to be effective.

[0051] In some alternative implementations, the normalized B-spline coefficients output after reinforcement learning convergence can be denormalized to actual physical thickness control parameters through linear mapping, while the normalized fiber layup angles are restored to true angle values ​​according to the original domain range. A three-dimensional model of the composite laminate flywheel is constructed in finite element software, and high-order solid elements such as Solid186 that support layered modeling and interlayer shear behavior capture are used for mesh generation to ensure local densification in areas with abrupt thickness changes or high gradients. The flywheel cross-sectional geometry is defined layer by layer according to the thickness distribution function obtained by denormalization, and an alternating layup sequence is set according to the optimal fiber layup angle, such as 0-degree and 90-degree interleaving or ±θ symmetrical layup.

[0052] In some optional implementations, a fixed constraint is applied at the inner diameter of the flywheel, and a rated angular velocity load is applied to the overall structure. The radial stress field, circumferential stress field, interlayer shear stress distribution, total mass, and moment of inertia of the entire flywheel are obtained by solving the problem, and then the energy storage density per unit mass is calculated. The four key indicators obtained from the finite element simulation are compared point by point with the prediction results of the physical information neural network at the same radial position. If the radial stress error does not exceed 3%, the circumferential stress error does not exceed 3%, the interlayer shear stress error does not exceed 5%, and the energy storage density per unit mass error does not exceed 2%, then the optimization result is deemed valid.

[0053] The verification process uses material properties and boundary conditions consistent with those generated from the training data to ensure a unified comparison benchmark. The inverse normalization formula strictly corresponds to the normalization strategy in the training phase to avoid scale bias from introducing verification errors.

[0054] Example 2: Based on the above examples, a specific implementation of a flywheel energy storage system variable thickness flywheel profile optimization method can be designed as follows: Considering that the flywheel of the flywheel energy storage system is made of composite materials, which have the characteristics of high specific strength and anisotropy, an objective function for optimizing the profile of the variable thickness flywheel of the flywheel energy storage system is established.

[0055] In an optional implementation, the objective function can be set to include objective 1 and objective 2, where objective 1 is set to maximize the energy storage density per unit mass. (Unit: J / kg), the calculation formula is as follows: The low density of composite materials allows them to achieve a higher moment of inertia for the same mass. The focus is on maximizing through profile design. Simultaneously, the directional properties of composite materials are utilized to unlock rotational speed potential. Among them, the total mass... Moment of inertia Rated angular velocity The high allowable stress of composite materials significantly improves and further amplifies the performance. The advantages.

[0056] Furthermore, objective 2 is set as minimizing the maximum centrifugal stress. Considering the anisotropy of composite materials, the radial stress is borne by the matrix, which often becomes a constraint bottleneck. Profile design needs to balance the biaxial stress through thickness distribution adjustment. The fiber directions are controlled as circumferential stresses (labeled as...). Direction) and transverse radial stress (marked as) (Direction), to avoid fiber breakage or interlayer delamination, allowable circumferential stress Radial allowable stress The optimization conditions are met. and .

[0057] Furthermore, the target priority weights are adapted to the characteristics of composite materials to maximize the energy storage density per unit mass. The weighting coefficient is Minimize maximum centrifugal stress The weighting coefficient is (Emphasis on stress control, with a focus on radial stress constraint). If considering the high-speed potential of composite materials, the focus is on increasing energy storage density. If radial stress constraint is considered, the stress control will increase. The weight calculation formula is as follows: It should be noted that subsequent evaluation optimizations incorporate biaxial stresses separately through a penalty function to avoid masking the risk of anisotropy by a single stress index.

[0058] Furthermore, a constraint system is established for the geometric boundaries, anisotropic material properties, and orthotropic mechanical laws of the composite flywheel.

[0059] In one alternative implementation, the geometric constraints focus on adapting the thickness range to the composite molding process. This includes the inner diameter. , outer diameter Thickness function .

[0060] In one alternative implementation, the material parameters take into account orthotropic properties, with the principal direction being... To elastic modulus Secondary direction To elastic modulus ; Towards deformation pair The influence of direction on the principal Poisson ratio Poisson's ratio Interlaminar shear properties and shear modulus ;density ; To stretch, The allowable stresses for tension and interlaminar shear are respectively , , .

[0061] In one alternative implementation, the physical constraints are based on the orthotropic elasticity reconstructed governing equations, taking into account... , and , The cross-coupling is a key characteristic that distinguishes the stress field of composite materials from that of metals.

[0062] The equilibrium equations take into account the coupling of radial and circumferential forces. Geometric equations , In some alternative implementations, the constitutive equations are modified to an orthotropic form. Furthermore, considering the constraints of the laminated structure, the fiber layup angle is taken into account. right , To mitigate the impact of optimization, ensure that the results are adapted to the actual molding process.

[0063] Furthermore, the laminated structure and molding process characteristics of the composite flywheel were precisely parameterized using a third-order B-spline function to achieve accurate parameterization of the thickness distribution. The thickness function is defined as follows: ( ),in The basis functions are 3rd order B-spline functions, and the nodal vectors are optimized based on the stress gradient characteristics of the composite material. Considering the radial stress of the composite flywheel within its inner diameter... The gradient is greatest in the vicinity, while the circumferential stress is at the outer edge. The nearest neighbor is more significant, so the node vector is set to The basis functions have higher interpolation accuracy near the inner diameter, providing a more nuanced description of thickness variations in that region. Optimizing the variables of the B-spline coefficients... The value range is adapted to composite material lamination processes. Minimum thickness Maximum thickness This constitutes a 4-dimensional continuous design space. To ensure the continuity of the laminated structure, the difference between adjacent coefficients needs to be constrained to avoid abrupt changes in the layup angle caused by sudden changes in thickness.

[0064] Furthermore, based on a parameterized thickness function and considering the low-density characteristics of composite materials, an analytical relationship between the moment of inertia and optimization variables is established, providing accurate input for energy storage density calculation. The formula for calculating the moment of inertia is as follows: The integral is decomposed into analytical integrals for each interval, and the 20-point Gaussian integration method is used for each interval, with the integration accuracy controlled within a certain range. Within. ,in It is a constant related to the basis function integral, density, and geometric dimensions, enabling a fast mapping between the moment of inertia and the optimization variables.

[0065] Furthermore, considering the influence of the orthogonal anisotropy of composite materials on the mechanical response, a high-quality dataset containing profile parameters, radial coordinates, and anisotropic stress / displacement is generated. Sample sampling employs an improved Latin hypercube sampling method, taking into account that the radial stress of composite materials is more sensitive to thickness variations. The sample density was increased within the sampling interval, resulting in a total of 60 sets of optimized variable samples. This ensures coverage of typical profile features such as "thin inner diameter, thick outer edge," "uniform thickness," and "local thickening." Finite element simulation models the composite material based on its properties, defining the laminated section and setting the material properties to orthogonal anisotropy; a rated angular velocity is applied. Then, the stress field and displacement field are obtained by solving. Discrete data sampling focuses on regions with significant stress gradients, in the radial coordinate... The above points are discrete, and the radial stress is output at each point. Circumferential stress radial displacement and interlaminar shear stress .

[0066] Furthermore, the final dataset format is as follows: It contains sufficient anisotropic mechanical information, improves data representativeness by encrypting key region sampling, and provides training samples for the physical constraints of PINN fused composite materials.

[0067] Furthermore, a PINN network structure is constructed to accurately capture the stress field characteristics of the orthotropic flywheel of composite materials. The network input and output dimensions and hidden layer structure are optimized to ensure that the model can both learn data features and be compatible with the physical laws of composite materials.

[0068] Furthermore, the input layer dimension of the PINN network is expanded to 6 dimensions, extending the original radial coordinates... With profile parameters Based on this, taking into account the ply angle Directly affects the equivalent elastic parameters of composite materials ( Increase the angle parameter between the principal direction and the radial direction of the fiber. The hidden layers employ a 4-layer fully connected structure. The first three layers each contain 128 neurons, with the Tanh activation function; the fourth layer contains 64 neurons, with the Swish activation function. This enhances the ability to fit interlaminar shear stress. The output layer is expanded to 4D, with radial stress... Circumferential stress radial displacement Interlaminar shear stress This enables end-to-end prediction of multiple stress components in composite flywheels, covering their main failure modes.

[0069] Furthermore, we define a PINN loss function that incorporates the physical laws of composite materials.

[0070] The total loss function is The design of each item is as follows: (1) Data loss The error includes all stress components, and the formula is: in (Total dataset) To label the interlaminar shear stress values ​​in the finite element simulation, by adding... The error term ensures the model's accuracy in learning the interlaminar properties of composite materials.

[0071] (2) Physical loss Based on the reconstruction of orthotropic elasticity, the equilibrium equations and the modified constitutive equations are satisfied. The form of the equilibrium equations remains unchanged. The constitutive equation is in the orthotropic form of the composite material: ( , ).

[0072] The physical loss is achieved by minimizing the residual of the above equation, as follows: (3) Boundary loss : Inner diameter fixed Outer diameter is free. Boundary constraints of interlaminar shear stress The formula is: (4) Shear stress penalty term To address the low interlaminar shear strength of composite materials, a penalty for excessive shear stress is added to prevent the model from underestimating interlaminar risk. The weight settings for the loss term were obtained through sensitivity analysis. , , To balance the fitting accuracy of various mechanical properties.

[0073] Furthermore, the PINN training for composite material data employs the AdamW optimizer with a decay rate of... , Weight decay coefficient Compare the PINN predictions with the finite element results, and stop when the following requirements are met to ensure that the model has fully learned the anisotropic characteristics.

[0074] Radial stress error ; Circumferential stress error ; Interlaminar shear stress error ; Displacement error .

[0075] Furthermore, a multi-objective PPO optimization algorithm is established, which sets optimization objectives for composite flywheels that take into account multiple stress components, energy storage, and processes. By enhancing the feature extraction and strategy expression capabilities of the PPO network, refined decision-making is achieved.

[0076] Furthermore, state space Expanded to 5 dimensions, based on the original 4-dimensional normalized profile parameters ( Based on this, a normalized value for the fiber layup angle is added. ( , For the actual ply angle, such as Action space It is refined into a 5-dimensional continuous space, including 4 profile parameter adjustment amounts. With a ply angle fine-tuning amount Reward function Incorporating multi-directional stress constraints, the penalty function is modified as follows: in: The energy storage density benchmark value for a composite flywheel of equal thickness; , , These represent the maximum radial stress, circumferential stress, and interlaminar shear stress predicted by PINN, and their allowable values ​​are respectively... , , ; For the compound penalty term, the weight is set to... , , , , It emphasizes the control of radial stress and interlaminar shear.

[0077] Furthermore, a PPO algorithm adapted to multi-objective composite materials is constructed. (Policy network) The input is a 5-dimensional state. The process involves a 4-layer fully connected network, with residual connections added between the first two layers and the last two layers. The output layer is divided into two branches: one branch outputs the average of the five actions. (Using Tanh activation), another branch outputs the standard deviation of 5 actions. (Activated using Softplus), the final action distribution is as follows Value Network Input status The network is then divided into two channels: The energy storage-thickness channel takes 4-dimensional thickness parameters (c0, c1, c2, c3) as input, employs a 32→16 fully connected structure, uses ReLU as the activation function, and outputs a 16-dimensional feature vector representing the "energy storage potential score corresponding to the thickness parameters." The stress-angle channel takes 1-dimensional layup angle (α) as input and 4-dimensional thickness parameters (c0, c1, c2, c3) as input. It also employs a 32→16 fully connected structure with the activation function Swish(x·sigmoid(x)). The network outputs a 16-dimensional feature vector representing the "stress safety score corresponding to the thickness and angle."

[0078] Furthermore, a multi-objective PPO optimization model is trained. He initialization based on an empirical design of a composite flywheel is used to reduce ineffective exploration. Starting from the initial state, actions are executed. Post-update status Obtain in one go through the PINN model , , , Substitute into the reward function to calculate To strengthen the focus on interlaminar shear stress, when When, adjustments are triggered Prioritize mitigating inter-layer risks. Trajectory data is stored as... When the average reward changes over 10 consecutive rounds In 95% of the sampling states , , Training has stopped.

[0079] Furthermore, the PPO was used to solve for the design parameters of the variable thickness flywheel profile of the flywheel energy storage system, and the effectiveness of the optimization results was ensured through dual verification by performance indicators and finite element simulation.

[0080] Furthermore, in the training of the convergent PPO policy network, the state with the highest reward value is selected. It contains 5-dimensional information: The complete parameters are restored through inverse normalization. The inverse normalization formula for the B-spline coefficients is: ( The optimal thickness coefficient is obtained; the inverse normalization formula for the ply angle is... ( (Using the angle normalized value in the state) to determine the optimal fiber orientation. Substitute into the 3rd order B-spline function This yields the optimal thickness distribution along the radial direction; combined with Together with the lamination process, a complete composite material flywheel design scheme is formed.

[0081] Furthermore, calculate the optimal moment of inertia. Combined with rated angular velocity With total mass The energy storage density per unit mass was obtained. The complete stress field is predicted using the PINN model, and three key stress indices are extracted: maximum radial stress. Maximum circumferential stress Maximum interlaminar shear stress If the verification criteria are met, the verification is considered successful. These criteria include energy storage performance indicators and stress safety margin indicators.

[0082] Furthermore, finite element simulation verification of the composite laminate model was performed. A high-precision composite laminate model was established in ANSYS using Solid186 elements and optimal design. Mesh the material to ensure mesh continuity at interlayer interfaces; define material properties according to orthotropic anisotropy, and set the layup angle to... and Alternating. After applying the rated angular velocity, the stress field (including interlayer shear stress) and energy storage parameters are obtained by solving. The simulation results are compared with the theoretical predictions, and the error is considered acceptable if it meets the following requirements.

[0083] Radial stress error Threshold; Circumferential stress error ; Interlaminar shear stress error ; Energy storage error .

[0084] Example 3, referring to Figure 2 This embodiment also provides a variable thickness flywheel profile optimization system for flywheel energy storage systems, comprising: The model building module is used to build a bi-objective optimization model with the optimization objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress. The constraint establishment module is used to construct a constraint condition system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; The parameterization module is used to parameterize the thickness distribution of the flywheel along the radial direction using a third-order B-spline function. The B-spline coefficients are used as optimization variables, and combined with the fiber layup angle to form a design parameter space and establish a parameterized model. The dataset acquisition module is used to generate a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress based on the parametric model and through finite element simulation. The neural network construction module is used to build a physical information neural network and train the physical information neural network using a dataset until its prediction results meet the preset stress and displacement error limits. The optimization module is used to construct the near-end policy optimization algorithm. During the policy iteration process of the near-end policy optimization algorithm, the trained physical information neural network is called to provide stress and energy storage feedback in real time until the policy converges and outputs the optimal B-spline coefficient and the optimal fiber layup angle.

[0085] The above-mentioned unit modules can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0086] This embodiment also provides an electronic device, which can be a terminal, and its internal structure diagram can be as follows. Figure 2 As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for optimizing the profile of a flywheel energy storage system with variable thickness. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the device's casing, or an external keyboard, touchpad, or mouse.

[0087] This embodiment also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it performs the following steps: A dual-objective optimization model is established with the objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress. Construct a constraint system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; The thickness distribution of the flywheel along the radial direction is parameterized using a third-order B-spline function. The B-spline coefficients are used as optimization variables, and a design parameter space is formed by combining the fiber layup angle to establish a parameterized model. Based on the parametric model, a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress is generated through finite element simulation. Construct a physical information neural network and train it using a dataset until its prediction results meet the preset stress and displacement error limits. A proximal policy optimization algorithm is constructed. During the policy iteration process of the proximal policy optimization algorithm, the trained physical information neural network is called to provide stress and energy storage feedback in real time until the policy converges and the optimal B-spline coefficient and the optimal fiber layup angle are output.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0089] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0090] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for optimizing the profile of a flywheel with variable thickness in a flywheel energy storage system, characterized in that, include: A dual-objective optimization model is established with the objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress. Construct a constraint system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; The thickness distribution of the flywheel along the radial direction is parameterized using a third-order B-spline function. The B-spline coefficients are used as optimization variables, and a design parameter space is formed by combining the fiber layup angle to establish a parameterized model. Based on the parametric model, a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress is generated through finite element simulation. Construct a physical information neural network, and train the physical information neural network using the dataset until its prediction results meet the preset stress and displacement error limits; A proximal policy optimization algorithm is constructed. During the policy iteration process of the proximal policy optimization algorithm, the trained physical information neural network is called to provide stress and energy storage feedback in real time until the policy converges and the optimal B-spline coefficient and the optimal fiber layup angle are output.

2. The method for optimizing the profile of a flywheel energy storage system with variable thickness as described in claim 1, characterized in that, The constraint system includes: Set the minimum and maximum values ​​of the flywheel's inner diameter, outer diameter, and thickness function; Set the elastic modulus, principal Poisson's ratio, secondary Poisson's ratio, shear modulus, and density of the composite material in the principal and secondary directions, as well as the allowable stresses for tensile stress in the principal direction, tensile stress in the secondary direction, and interlaminar shear. Based on orthotropic elasticity, the equilibrium equations of radial and circumferential forces, strain-displacement geometric relations, and stress-strain constitutive relations are established. A continuity constraint is set on the fiber layup angle to limit drastic changes in the layup angle caused by abrupt changes in thickness in adjacent areas.

3. The method for optimizing the profile of a flywheel energy storage system with variable thickness as described in claim 2, characterized in that, The parameterization of the flywheel thickness distribution using a third-order B-spline function includes: Based on the characteristics of a large radial stress gradient near the inner diameter of the flywheel and a significant circumferential stress near the outer edge, a non-uniformly distributed node vector is set. The thickness function is represented as a linear combination of B-spline basis functions and coefficients, where the coefficients are treated as continuous optimization variables. An upper limit constraint on the difference between adjacent B-spline coefficients is applied to ensure that the thickness varies smoothly in the radial direction.

4. The method for optimizing the profile of a flywheel energy storage system with variable thickness as described in claim 3, characterized in that, The dataset includes: An improved Latin hypercube sampling method is used to extract sample points in the design space composed of B-spline coefficients. Increase the sample density in the region near the inner diameter of the flywheel; For each sample point, a finite element model of the composite laminate structure is established, the orthogonal anisotropic material properties and laminate section are defined, a rated angular velocity load is applied, and the stress field and displacement field are obtained by solving. Multiple discrete locations are selected on the radial coordinate, and the radial stress, circumferential stress, radial displacement, and interlaminar shear stress at each location are extracted to form training samples.

5. The method for optimizing the profile of a flywheel energy storage system with variable thickness as described in claim 4, characterized in that, The training loss of the physical information neural network includes four terms: The first item is data loss, which measures the deviation between the network output and the finite element simulation results in each stress component and displacement. The second term is physical loss, which measures whether the network output satisfies the equilibrium equation and constitutive relation of orthotropic materials. The third term is boundary loss, which is used to force the satisfaction of the constraint conditions that the displacement at the inner diameter is zero, the stress at the outer diameter is free, and the interlayer shear stress is at the boundary. The fourth item is the shear penalty loss, which is used to impose an additional penalty when the interlaminar shear stress exceeds the allowable value; The physical information neural network takes radial coordinates, B-spline coefficients and fiber layup angle as inputs, and radial stress, circumferential stress, radial displacement and interlaminar shear stress as outputs. During the training process, orthogonal anisotropic elastic mechanical control equations, boundary conditions and interlaminar shear strength constraints are introduced as physical loss terms.

6. The method for optimizing the profile of a flywheel energy storage system with variable thickness as described in claim 5, characterized in that, The proximal policy optimization algorithm includes a policy network and a value network: The state space of the near-end strategy optimization algorithm consists of normalized B-spline coefficients and normalized fiber layup angles. The action space includes the adjustment amount of the B-spline coefficients and the fine-tuning amount of the fiber layup angles. The reward function is constructed based on the degree of improvement of the energy storage density per unit mass and whether the stress in each direction exceeds the corresponding allowable value. The policy network receives the current state, processes it through multiple fully connected layers, and outputs the mean and standard deviation of the actions to form a probability distribution of consecutive actions. The value network contains two parallel processing channels. The first channel receives the thickness parameters and evaluates their corresponding energy storage potential, while the second channel receives the thickness parameters and fiber layup angle and evaluates their corresponding stress safety. During the strategy update process, if the interlaminar shear stress predicted by the physical information neural network exceeds the allowable value, the action sampling logic is adjusted to prioritize actions that reduce the risk of interlaminar shear.

7. The method for optimizing the profile of a flywheel energy storage system with variable thickness as described in claim 6, characterized in that, It also includes a verification step: The optimal B-spline coefficients and optimal fiber layup angle obtained after convergence are inversely normalized to obtain the actual thickness distribution function and fiber layup angle. A composite laminate flywheel model was established in finite element software. The mesh was generated using an element type that adapted to the interlayer interface. The thickness distribution and the angle of the alternately laid fibers were set according to the inverse normalization results. Apply a rated angular velocity load and calculate the radial stress, circumferential stress, interlayer shear stress, and energy density per unit mass of the flywheel. The finite element calculation results are compared with the predicted values ​​of the physical information neural network. If all errors are within the preset tolerance range, the optimization results are confirmed to be effective.

8. A variable thickness flywheel profile optimization system for a flywheel energy storage system, using the method described in any one of claims 1 to 7, characterized in that, include: The model building module is used to build a bi-objective optimization model with the optimization objectives of maximizing energy storage density per unit mass and minimizing maximum centrifugal stress. The constraint establishment module is used to construct a constraint condition system that includes flywheel geometric boundaries, orthogonal anisotropic properties of composite materials, and process constraints of laminated structures; The parameterization module is used to parameterize the thickness distribution of the flywheel along the radial direction using a third-order B-spline function. The B-spline coefficients are used as optimization variables, and combined with the fiber layup angle to form a design parameter space and establish a parameterized model. The dataset acquisition module is used to generate a dataset containing design parameters, radial position and its corresponding radial stress, circumferential stress, radial displacement and interlayer shear stress based on the parametric model and through finite element simulation. The neural network construction module is used to construct a physical information neural network and train the physical information neural network using the dataset until its prediction results meet the preset stress and displacement error limits. The optimization module is used to construct the near-end policy optimization algorithm. During the policy iteration process of the near-end policy optimization algorithm, the trained physical information neural network is called to provide stress and energy storage feedback in real time until the policy converges and outputs the optimal B-spline coefficient and the optimal fiber layup angle.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the variable thickness flywheel profile optimization method for a flywheel energy storage system according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the variable thickness flywheel profile optimization method for a flywheel energy storage system according to any one of claims 1 to 7.

Images