Direct-current cable terminal structure optimization method and device based on double-layer integrated stacking optimization proxy model, and medium

Abstract

The invention discloses a direct-current cable terminal structure optimization method and device based on a double-layer integrated stacking optimization proxy model and a medium. The method comprises the following steps: (1) carrying out trapezoid-chamfer parametric modeling on an outer side curve of an epoxy sleeve, and constructing a training sample by Latin hypercube sampling; (2) constructing a'base learner-meta learner 'double-layer integrated stacking agent model, and synchronously optimizing model configuration and hyper-parameters by using an improved sparrow search algorithm; (3) driving ISSA (International Standard Standard Architecture) to quickly optimize by using the trained proxy model to obtain an optimal geometric curve; and (4) through finite element-stream theory-thermal shock joint verification, it is confirmed that the maximum electric field intensity of the optimized terminal is reduced, the initial discharge voltage is increased, and the mechanical reliability meets the long-term operation requirement. The method solves the problem that a traditional agent model is insufficient in precision and insufficient in insulation margin of a gas-solid interface of a 550kV direct-current cable terminal, and is widely applied to localization design of + / -550kV GIS cable terminals.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-voltage direct current transmission equipment, and particularly relates to a DC cable terminal structure optimization method, equipment and medium based on a double-layer integrated stacking optimization agent model. BACKGROUND

[0002] With the large-scale construction of ±550kV and above DC transmission projects, the GIS cable terminal as the key interface between the submarine cable and the converter station is prone to accumulate surface charge under long-term unipolar DC voltage, resulting in gas-solid interface electric field distortion, triggering surface flashover or even breakdown. In addition, the temperature gradient caused by conductor heating will further change the material conductivity distribution, exacerbating the electric field unevenness. Traditional finite element global optimization is time-consuming due to multi-field coupling and high-dimensional parameters; a single agent model is difficult to balance accuracy and generalization ability. Therefore, there is an urgent need for a structure optimization method that balances computational efficiency and prediction accuracy to improve the insulation margin and operation reliability of the epoxy sleeve. SUMMARY

[0003] The purpose of the present application is to provide a DC cable terminal structure optimization method, equipment and medium based on a double-layer integrated stacking optimization agent model, to solve the problems of electric field concentration, insufficient insulation margin and low optimization efficiency in the prior art.

[0004] To achieve the above-mentioned purpose, the following technical solutions are adopted in the present application: a DC cable terminal structure optimization method based on a double-layer integrated stacking optimization agent model, comprising the following steps:

[0005] S1 parameterized modeling:

[0006] S1-1 parameterize the structure curve outside the epoxy sleeve with a trapezoidal curve and a chamfer to establish a geometric model;

[0007] S1-2 use Latin hypercube sampling algorithm to extract samples in the structure parameter space of the geometric model to form a sample data set;

[0008] S2 double-layer integrated stacking agent model construction:

[0009] S2-1 formalize the model selection problem as a decision variable

[0010] M=[m1,m2,…,mNm,mNm+1],

[0011] Where m1~mNm are binary variables indicating whether to include the corresponding base learner in the stack; mNm+1 is an integer variable representing the meta-learner number;

[0012] S2-2 establish a double-layer optimization framework:

[0013] Upper layer: use k-fold cross-validation average RMSE as fitness, solve optimal M by ISSA;

[0014] Lower layer: after fixing M, use ISSA to optimize the hyperparameters of each base / meta learner again, and return the minimum RMSE to the upper layer;

[0015] S2-3 introduces

[0016] Logistic chaotic initialization: x n+1 =α L x n (1-x n )

[0017] In the formula: α L is the Logistic control parameter, and α L ∈(0,4], x n ∈(0,1);

[0018] Binary transfer function: map continuous positions to 0 / 1, realize the self-adaptation of the sparrow search algorithm to discrete space;

[0019] S2-4 trains the double-layer stacked proxy model with the optimal configuration obtained in steps S2-2 and S2-3, and establishes the mapping between the structure parameters and the objective function;

[0020] S3 structure optimization:

[0021] S3-1 embeds the trained proxy model into the main loop of ISSA to quickly evaluate the candidate structure;

[0022] S3-2 outputs the optimal geometric parameters after iterative convergence;

[0023] S4 verification closed loop:

[0024] S4-1 substitutes the optimal geometric parameters into the finite element model to calculate the gas-solid interface electric field distribution;

[0025] S4-2 calculates the discharge starting voltage by using the streamer theory;

[0026] S4-3 performs thermal-mechanical coupling stress checking to confirm the mechanical integrity of the epoxy sleeve under thermal shock.

[0027] Further, the base learner includes at least three of extreme gradient boosting tree XGBoost, support vector regression SVR, random forest RF, and light gradient boosting machine LightGBM; and the meta learner is a random forest or a ridge regression.

[0028] Further, in the trapezoidal-chamfer parameterization of step S1-1, the chamfer radii ra and rb are independently adjustable, wherein the value range of the upper chamfer radius ra is determined by the maximum chamfer size allowed by the structure.

[0029] Further, in the k-fold cross-validation of step S2-2, k = 5.

[0030] Further, the k-fold cross-validation score is introduced into the objective function, and optimization is performed

[0031]

[0032] In the formula, K S is the number of k-fold cross-validation; RMSE is a regression evaluation index, i.e., the root mean square error;

[0033] is the predicted value, y i is the actual value.

[0034] Further, when the applied voltage meets the streamer criterion in the streamer theory calculation of the discharge starting voltage of step S4-2, discharge occurs.

[0035]

[0036] In the formula, for SF6 gas, K is 18, x cr is the path along the electric field line;

[0037] For SF6 gas, considering the relationship between gas density and dielectric strength, (α-η) can be calculated by the following formula:

[0038]

[0039] The present application provides an electronic device, comprising:

[0040] A memory storing a computer program;

[0041] A processor, when executing the computer program, realizes the steps of the method of any one of the above.

[0042] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, realizes the steps of the method of any one of the above.

[0043] Advantages

[0044] The present application can significantly improve the electric field distribution on the gas-solid interface of ±550kV GIS cable terminal, especially in the case of large temperature gradient change. By reducing the maximum electric field strength, the insulation margin problem caused by electric field distortion is effectively alleviated, and better overall performance in offshore wind power applications is ensured.

[0045] Under the typical working condition of ±550 kV, the maximum electric field strength is reduced by 12.8%, and the discharge inception voltage is increased by 15%. The improvement of electric field distribution directly leads to the increase of discharge voltage, thereby reducing the risk of breakdown at the gas-solid interface. This voltage increase indicates that this method has the potential to enhance the insulation performance, especially under high temperature gradient conditions, which traditional design may not be able to effectively cope with.

[0046] The present application combines two-layer stacked integrated optimization agent model and improved sparrow search algorithm (ISSA), greatly reducing the computational cost and time. The acceleration of the optimization process reflects a significant reduction in computing time, with traditional optimization methods requiring more than 7800 hours, while the optimization time of this new technology is only 293.48 hours. This improvement in computational efficiency makes the design process more efficient and feasible in practical applications.

[0047] The first principal stress of the optimized epoxy sleeve in the thermal shock test only rises by 3%, meeting the mechanical reliability requirements. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1550 kV DC cable terminal two-dimensional axisymmetric geometric model.

[0049] Figure 2 Electric field distribution at the gas-solid interface before optimization.

[0050] Figure 3 Parametric diagram of trapezoidal-chamfer outside the epoxy sleeve.

[0051] Figure 4 Influence of different chamfer radii on the maximum electric field strength.

[0052] Figure 5 Flowchart of double-layer integrated stacked agent model optimization.

[0053] Figure 6 Comparison of different agent model accuracy.

[0054] Figure 7 Electric field distribution at the gas-solid interface after optimization.

[0055] Figure 8 Comparison of discharge inception voltage before and after optimization.

[0056] Figure 9 Comparison of thermal-mechanical stress before and after optimization.

[0057] Figure 10 Global framework for structural optimization of DC cable terminal. DETAILED DESCRIPTION

[0058] The present application will be further described below in conjunction with the drawings and examples.

[0059] The 550kV DC cable terminal structure optimization method based on the double-layer integrated stack optimization agent model of the application will be further described below in combination with the drawings and specific embodiments.

[0060] Step S1: The geometric model of the ±550kV DC submarine cable terminal is as shown in Figure 1 The terminal is vertically placed, and the basic structure mainly consists of five parts: cable body, preformed stress cone rubber, epoxy sleeve, shell and flange. In order to improve the solving speed while maintaining the solving accuracy, the application adopts a two-dimensional axisymmetric model for simulation. In addition, in order to simulate the real operating environment of the cable terminal, the shell temperature is set to 15℃, and the conductor is 75, 55, 35, 15℃ to simulate the temperature difference condition under different load currents. The rest of the terminal is filled with SF6 gas with a pressure of 0.5MPa, and the fluid property is set to turbulent flow. The model considers the heat transfer between each part under the condition of gas flow. The conductor is applied with +550kV high voltage, and the flange, insulation shield and infinity are set to ground. The heat flow field calculation comprehensively considers heat conduction, heat convection and heat radiation and other heat transfer mechanisms to simulate the temperature distribution of the DC cable terminal under actual operating conditions. The electric field is calculated based on the charge transport equation, and the electric field distribution after 50 hours of charge accumulation is used as the steady-state electric field calculation result.

[0061] The electric field distribution of the epoxy sleeve gas-solid interface under different temperature gradients before optimization is as shown in Figure 2 Under the condition of constant room temperature, the electric field concentration phenomenon occurs at the upper chamfer of the epoxy sleeve, and the maximum electric field strength is 6.4kV / mm, which is much higher than the average electric field strength. With the continuous rise of temperature difference, the maximum surface electric field moves to the abdomen of the epoxy sleeve, and the electric field strength shows a gradually increasing trend, and the maximum electric field strength rises to 7.0kV / mm, which is located in the abdomen of the epoxy sleeve. This may be due to the decrease of electrical conductivity on the outside of the epoxy sleeve with the increase of distance from the conductor, and the uneven distribution of material electrical conductivity caused by the electrical conductivity gradient leads to the intensification of charge accumulation, and then induces electric field distortion.

[0062] Step S2: The application first selects A1 (upper side of the metal insert) and A2 (flange side) as the reference reference points, and B3, B4, B5 and B6 coordinates as the coordinates to be adjusted. From Figure 3 It can be seen from the above table that B3 and B4 and B5 and B6 have the same coordinates on the y-axis, and B3 and A1 and B5 and A2 have the same coordinates on the x-axis. Therefore, the application defines x a , x b , y a , y b to represent the relationship between the coordinate points, and x a , x b , ya , y b as variables to be optimized.

[0063] x a = |x B4 -x B3 |

[0064] x b = |x B6 -x B5 |

[0065] y a = |y B3 -y A1 |

[0066] y b = |y B5 -y A2 |

[0067] After determining the coordinate relationship of each point, chamfer optimization is needed to alleviate the influence of electric field distortion at the chamfer on the insulation performance. Therefore, the built-in function of the finite element software is used to set and adjust the upper and lower chamfer sizes, which are respectively defined as r a and r b , as shown in Figure 3 (b).

[0068] During the simulation calculation process, the present application notices that the chamfer configuration not only significantly affects the electric field distribution, but also causes interference to the quality and accuracy of mesh division when the chamfer is parameterized as a design variable in finite element analysis, which may further affect the training effect and prediction accuracy of the surrogate model. Therefore, in order to ensure the reliability and stability of the optimization model, the present application analyzes the influence of different chamfer parameter configurations on the maximum electric field distribution before constructing the structural optimization model.

[0069] As shown in Figure 4 , under the working condition of temperature difference ΔT = 60℃, with the increase of the upper chamfer radius r a , the maximum electric field strength E max gradually decreases and tends to be stable, while the lower chamfer radius r b has almost no effect on the electric field distribution. This shows that appropriately increasing the upper chamfer radius r a helps to effectively alleviate the problem of local electric field concentration. Based on this conclusion, the present application sets r a to the maximum value allowed by the finite element software, i.e. r a = x a , and r b is set to 20mm according to the existing engineering design, so as to simplify the model complexity and improve the simulation efficiency.

[0070] Step S3: design a proxy model based on a double-layer integrated stack optimization method to map the mapping relationship between the structure input parameters and the optimization target.

[0071] The construction method of the double-layer integrated stack optimization method proxy model is as follows

[0072] Step S301: define the learner selection configuration problem of the stack proxy model, assume that the decision variables M=[m1, m2, …, m Nm , m Nm+1 ] are defined, wherein the binary variables m1~m Nm (taking 1 indicates selecting the model as the base model, otherwise 0) are used to screen the best combination of base learners, and the integer variables m Nm+1 ∈[1, N m ] are used to select the optimal meta-learner.

[0073] Step S302: define a double-layer optimization framework, in the upper-layer optimization, select the optimal stack configuration, and take the value of the objective function as the fitness function of the ISSA to calculate, in the lower-layer optimization process, optimize the hyperparameter configuration of each learner through the ISSA, calculate the corresponding root mean square error, and use the obtained optimal root mean square error as the fitness value of the upper-layer optimization, so as to realize the closed-loop iteration of the upper and lower layers.

[0074] Step S303: introduce the K-fold cross-validation score into the objective function for optimization, and the specific description is as follows:

[0075]

[0076]

[0077] In the formula, K S is the number of k-fold cross-validation, which is set to 5 in this study; RMSE is a regression evaluation index, i.e., the root mean square error. is the predicted value, y i is the actual value. The smaller the value of RMSE, the better the performance of the prediction model.

[0078] Step S304: improve the SSA, introduce a binary transfer function to make the SSA adapt to binary optimization tasks, and introduce a chaotic initialization strategy to improve the quality of the initial solution set of the SSA, and the specific description is as follows:

[0079] The binary transfer function calculation formula is:

[0080]

[0081] Wherein, represents the position of sparrow i at the iteration number t, T s [Xi t r is the conversion value according to the position update, r d is a random number in the range [0, 1].

[0082] The mathematical expression of the chaos initialization strategy is:

[0083] x n+1 = a L x n (1-a n )

[0084] In the formula: a L is the Logistic control parameter, and a L ∈(0, 4]; x n ∈(0, 1).

[0085] Step S305: Based on the objective function set in step S203, the model developed in steps S201 and S202 is used as a surrogate model, and the optimization strategy proposed in step S204 is used for optimization, so as to realize accurate prediction of the mapping relationship between the input structure parameters and the to-be-optimized objective function. The optimization process is as shown in Figure 5 .

[0086] Step S4: After completing the training of the surrogate model, the ISSA algorithm is iteratively optimized with the trained surrogate model to obtain the optimal structure parameters.

[0087] Table 1 shows the comparison of calculation efficiency between using the surrogate model and directly iterating the calculation with Comsol through the optimization algorithm. The average time of each finite element calculation is about 276.39 seconds.

[0088] If the surrogate model is not used, finite element calculation is required for each sample in the optimization algorithm.

[0089] In contrast, when using the surrogate model proposed in the present application, only 3750 sample points used to train the double-layer stacked surrogate model need to be calculated by finite element calculation, which significantly reduces the time required for finite element calculation and greatly improves the calculation efficiency.

[0090] Table 1 Comparison of optimization time of different optimization methods

[0091]

[0092] The regression accuracy of the stacking model in Table 2 and the regression accuracy of the stacking model in the literature in Table 2 are compared. It is worth noting that in order to ensure the effectiveness of the comparison, the hyperparameter configuration of the model mentioned in the application is trained by the sparrow optimization algorithm mentioned in the application, and the hyperparameters of the models in Table 4 are referred to the hyperparameters mentioned in the corresponding literature. The root mean square error and the mean absolute error of each algorithm are as shown in Table 2. Figure 6 The following conclusions can be drawn through comparative analysis:

[0093] Table 2 Optimal stacking configuration and hyperparameters

[0094]

[0095] 1) The structural parameters selected in the application have good representativeness, and all algorithms can complete the proxy regression of the terminal structure optimization.

[0096] 2) As can be seen from the figure, compared with the traditional mathematical algorithm, the machine learning method and the neural network method have better performance and are more suitable for the application scenario proposed in the application.

[0097] 3) The double-layer integrated stacking proxy optimization model proposed has the best performance among all algorithms, has lower MAE and RMSE, and it is worth noting that although CNN shows better proxy performance than SVM, the stacking optimization result still selects SVM as the base model, which fully illustrates the effectiveness of the stacking integration method of heterogeneous models, provides better flexibility, enhanced generalization and stronger ability to solve complex practical problems for the proxy model.

[0098] 4) Compared with the prediction performance of the stacking model in the reference literature, the proxy model has higher prediction accuracy, which further proves that it is necessary to configure the corresponding stacking model and optimize the hyperparameters for a specific engineering problem.

[0099] By improving the sparrow optimization algorithm, the optimal four structural parameter sizes are obtained as x a = 20.21 mm, x b = 23.87 mm, y a = 20.11 mm, and y b = 175.64 mm. Under the condition of a temperature difference of 60℃, the maximum electric field mode value of the gas-solid interface is 6.13kV / mm. Considering the actual manufacturing capacity, the optimized structural parameter sizes are rounded and listed in Table 3.

[0100] Table 3 Comparison of structural parameters before and after optimization

[0101]

[0102] Figure 7 It is shown that the optimized DC cable terminal gas-solid interface electric field distribution, it can be seen from the structure curve, the overall oblique arc increases, and the upper chamfer is closer to the metal insert, and the overall length of the terminal presents a downward trend. In addition, it can be seen that under different temperature gradients, the gas-solid interface electric field modulus decreases, and the gas-solid interface electric field modulus decreases with the increase of temperature gradient. The electric field strength of the optimized structure under different temperature gradients has good performance.

[0103] Step S5: The discharge inception voltage of the steady-state DC cable terminal gas-solid interface under different temperature gradients is evaluated by using the streamer theory to further verify the effectiveness of the structure optimization. The size of the discharge inception voltage depends on two aspects: the maximum value of the electric field around the gas-solid interface and the range of the high-intensity electric field region. The maximum value of the electric field is closely related to the discharge inception, while the range of the high-intensity electric field region affects the propagation of the streamer. The development of the streamer depends on whether the number of charges in the high-intensity electric field meets the streamer discharge standard. When the applied voltage meets the streamer standard, discharge occurs.

[0104]

[0105] In the formula, for SF6 gas, K is 18, x cr is along the electric field line path.

[0106] For SF6 gas, considering the relationship between gas density and dielectric strength, (α-η) can be calculated by the following formula:

[0107]

[0108] Figure 8 It is shown that the optimized structure under different temperature gradients before and after optimization, it can be seen that after optimization, under the conditions of temperature gradient of 0℃, 20℃, 40℃, 60℃, the starting discharge voltage increases by 9.1%, 8.9%, 12%, 15.1% respectively. The increase of the starting discharge voltage indicates that the optimized DC submarine cable terminal gas-solid interface has stronger insulation performance. In addition, with the increase of temperature gradient, the dielectric strength of the gas-solid interface gradually decreases, and the starting discharge voltage of the gas-solid interface continuously decreases, while the starting discharge voltage of the optimized structure under large temperature difference condition is more obvious, which fully proves the high strength insulation performance of the optimized structure under large temperature difference condition, and proves the effectiveness of the method.

[0109] Step S6:

[0110] Before the DC submarine cable terminal is put into operation, some samples need to be subjected to thermal shock test to detect the thermal stress of the epoxy insulation material under rapid temperature change from 5 DEG C to 95 DEG C. The thermal stress is caused by the thermal expansion mismatch between the conductor and the epoxy, which can cause internal cracking of the epoxy sleeve. In addition, residual stress further aggravates this risk, so it is essential to analyze the residual stress and thermal expansion stress at the same time. The calculation of residual stress considers the strain caused by thermal expansion and chemical shrinkage, while the thermal shock test only considers thermal strain. The formula is as follows:

[0111] {sigma} = [C] {epsilon c -epsilon T -epsilon r} + {sigma0}

[0112] epsilon T = alpha T (T-T ref )

[0113]

[0114] In the formula, sigma and sigma0 represent internal stress and initial stress respectively; epsilon T and epsilon r represent thermal strain and chemical shrinkage strain respectively; epsilon c is the total strain; [C] is the stiffness matrix. Alpha T represents the thermal expansion coefficient of the material, with a unit of 1 / K. T ref represents the reference temperature, with a unit of K. Delta alpha represents the curing degree, and the mechanical properties of the material are shown in Table 4.

[0115] Table 4 Mechanical property parameters

[0116]

[0117] Figure 9 The stress comparison of the structure thermal shock test before and after optimization is shown, and it can be seen that the maximum value of the first principal stress is located at the interface between the metal insert and the epoxy sleeve. And after coupling the residual stress, the maximum value of the first principal stress rises very obviously, so it is necessary to consider the residual stress for analysis. Figure 9 (a3) and (b3) show the distribution of the first principal stress of the structure before and after optimization, and it can be seen that compared with the structure before optimization, the first principal stress of the structure after optimization only rises by 3%, which meets the mechanical performance evaluation requirements, proving the effectiveness of the structure optimization method of the present application.

[0118] An electronic device comprises a memory and a processor connected in communication with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the model stacking configuration and hyperparameter optimization considering proxy model construction method of any one of the preceding aspects.

[0119] A computer storage medium having stored thereon computer instructions for causing a computer to perform the method of constructing a proxy model considering model stack configuration and hyperparameter optimization according to any one of the preceding.

[0120] Those skilled in the art will appreciate that embodiments of the present application can be supplied as methods, systems, or computer program products. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can be embodied in the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROMs, optical storage media, etc.) having computer usable program code embodied thereon. The procedures described in the embodiments of the present application can be implemented as a computer program product, using any of a variety of computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0121] The present application is described in reference to the flowcharts and / or block diagrams of the methods, apparatus (systems) and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions, which are executed via the processor of the computer or other programmable data processing apparatus, generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0122] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions means that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0123] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0124] Example 1

[0125] Step 1 Parametric modeling

[0126] With Figure 1 A two-dimensional axisymmetric model was established for the ±550 kV DC cable terminal. The conductor temperature was set to 75℃, 55℃, 35℃, 15℃, and the shell was 15℃, simulating different loads.

[0127] The outer curve of the epoxy sleeve was parameterized using a trapezoidal-chamfer ( Figure 3 ), with xa, xb, ya, yb, and chamfer radius ra, rb as design variables.

[0128] 3750 samples were extracted from the variable space by Latin hypercube sampling, and Comsol Multiphysics was used to calculate the steady-state electric field distribution to form the training set.

[0129] Step 2 Construction of double-layer integrated stacked proxy model

[0130] Base learner candidates: XGBoost, SVR, RF, LightGBM; meta-learner candidates: RF, Ridge.

[0131] Decision variable M as claimed in claim 1.

[0132] The upper ISSA uses the average RMSE of 5-fold cross-validation as the fitness, and the lower ISSA fine-tunes the hyperparameters of each learner.

[0133] ISSA improvement points:

[0134] Logistic chaos initialization: αL = 4;

[0135] Binary transfer function: sigmoid mapping + random threshold, realizing discrete space search.

[0136] Step 3 Structure optimization

[0137] Embed the trained proxy model into ISSA, population size 50, maximum iteration 100.

[0138] The optimal geometric parameters are: xa = 20mm, xb = 24mm, ya = 20mm, yb = 175mm, ra = 20mm, rb = 20mm (rounded in engineering).

[0139] Step 4 Verification closed loop

[0140] Finite element results: When ΔT = 60℃, the maximum electric field strength decreases from 7.0 kV / mm to 6.13 kV / mm (decrease of 12.8%).

[0141] Streamer theory calculation: discharge inception voltage increased by 15% ( Figure 8 ).

[0142] Thermo-mechanical coupling: thermal shock test increased the first principal stress by only 3% ( Figure 9 ), meeting the long-term operation requirements.

[0143] Example 2

[0144] An electronic device comprising a memory and a processor, the memory having stored therein a computer program implementing the method of the present application, the processor invoking the program to perform any of the steps of claims 1-6.

[0145] Example 3

[0146] A computer readable storage medium having stored therein a computer program implementing the method of the present application, the program when executed by a processor implementing any of the steps of claims 1-6.

[0147] The present application realizes the efficient and high-precision structure optimization of the 550kV DC cable terminal epoxy sleeve through the deep integration of the double-layer integrated stacked proxy model and ISSA, which is significantly superior to the prior art in terms of insulation performance, mechanical reliability and calculation efficiency, and has good engineering popularization value.

[0148] The above only describes the preferred embodiments of the present application, and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for optimizing the structure of DC cable terminations based on a two-layer integrated stacked optimization surrogate model, characterized in that, Includes the following steps: S1 parametric modeling: S1-1 uses trapezoidal curves and chamfers to parameterize the outer structural curve of the epoxy sleeve and establish a geometric model; S1-2 uses the Latin hypercube sampling algorithm to extract samples in the structural parameter space of the geometric model to form a sample dataset. S2 Two-Layer Integrated Stacked Proxy Model Construction: S2-1 formalizes the model selection problem into decision variables. M = [m1,m2,…,mNm,mNm+1], Where m1 to mNm are binary variables, indicating whether to include the corresponding base learner in the stack; mNm+1 is an integer variable, representing the meta-learner number; S2-2 establishes a two-layer optimization framework: Upper layer: Using the average RMSE of k-fold cross-validation as the fitness, the optimal M is solved using ISSA; Lower layer: After fixing M, ISSA is used again to optimize the hyperparameters of each basis / meta learner, and the minimum RMSE is returned to the upper layer; S2-3 was introduced in the ISSA. Logistic chaotic initialization: x n+1 =α L x n (1-x n ) In the formula: α L It is the control parameter for Logistic regression, and α L ∈(0,4],x n ∈(0,1); Binary transfer function: maps continuous positions to 0 / 1, enabling the sparrow search algorithm to adapt to discrete space; S2-4 uses the optimal configuration obtained in steps S2-2 and S2-3 to train the two-layer stacked proxy model and establish the mapping between structural parameters and objective function; S3 structure optimization: S3-1 embeds the trained surrogate model into the ISSA main loop to quickly evaluate candidate structures; After S3-2 iteration converges, the optimal geometric parameters are output. S4 verification closed loop: S4-1 substitutes the optimal geometric parameters into the finite element model to calculate the electric field distribution at the gas-solid interface; S4-2 uses the streamer theory to calculate the discharge initiation voltage; S4-3 performs thermo-mechanical coupling stress verification to confirm the mechanical integrity of the epoxy sleeve under thermal shock.

2. The method according to claim 1, characterized in that, The base learner includes at least three of the following: Extreme Gradient Boosting Tree (XGBoost), Support Vector Regression (SVR), Random Forest (RF), and Light Gradient Boosting Machine (LightGBM); the meta-learner is Random Forest or Ridge Regression.

3. The method according to claim 1, characterized in that, In the trapezoid-chamfer parameterization described in step S1-1, the chamfer radius ra and rb are independently adjustable, wherein the range of the upper chamfer radius ra is determined by the maximum chamfer size allowed by the structure.

4. The method according to claim 1, characterized in that, In step S2-2, k = 5 in the k-fold cross-validation.

5. The method according to claim 1, characterized in that, The k-fold cross-validation score is incorporated into the objective function for optimization. In the formula, K S It is the number of folds in the k-fold cross-validation; RMSE is the regression evaluation index, i.e., root mean square error. It is the predicted value, y i This is the actual value.

6. The method according to claim 1, characterized in that, In step S4-2, when calculating the discharge initiation voltage using the streamer theory, discharge occurs when the applied voltage meets the streamer criteria. In the formula, for SF6 gas, K is 18, x cr The path is along the electric field lines; For SF6 gas, considering the relationship between gas density and dielectric strength, (α-η) can be calculated using the following formula:

7. An electronic device, characterized in that, include: Memory, which stores computer programs; The processor, when executing the computer program, implements the steps of the method according to any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images