CN115455837A - Intelligent design optimization method for litz line winding - Google Patents
Intelligent design optimization method for litz line winding Download PDFInfo
- Publication number
- CN115455837A CN115455837A CN202211161074.4A CN202211161074A CN115455837A CN 115455837 A CN115455837 A CN 115455837A CN 202211161074 A CN202211161074 A CN 202211161074A CN 115455837 A CN115455837 A CN 115455837A
- Authority
- CN
- China
- Prior art keywords
- litz
- litz wire
- size
- wire winding
- line
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses an intelligent design optimization method for litz wire windings, which is characterized in that the alternating current resistance and the porosity of the litz wire windings are obtained through a neural network and hfss, and the alternating current resistance values of the litz wire windings under different materials, different numbers of turns, different litz wire winding radiuses and different frequencies are obtained through simulation by using hfss simulation software to carry out model construction on the litz wire windings, so that training data and test data used by a litz wire winding neural network model are completed. The adoption of manual meshing simplifies the solving step and time and ensures the accuracy. And (3) automatically training and testing by using a litz wire winding neural network model by using a particle swarm optimization algorithm to perfect the network model, and finally, quickly and accurately obtaining the required litz wire winding alternating current resistance by using the litz wire winding neural network model. The method can be used for quickly and accurately obtaining the litz wire winding alternating current resistance under different requirements, reduces the calculated amount and the working time, and greatly improves the efficiency.
Description
Technical Field
The invention relates to the technical field of wire design, in particular to an intelligent design optimization method for litz wire winding.
Background
In recent years, with the rapid development of human science and technology level and the continuous improvement of living standard, various electronic products, such as 5G mobile phones, tablet computers, wireless headsets, etc., are gradually integrated into the daily life of people, but the charging problem of these electronic devices is unavoidable. The traditional power supply generally adopts a wired mode, namely, a wire is used for supplying power to equipment, and the mode is simple and low in loss, and is the current mainstream power supply mode. However, as electronic products increase in life, a charger and a charging cable bring great cost and inconvenience, and in some special environments or special industries, a wired power supply mode has considerable limitations, such as: in an underground or underwater environment, damage to the wires may cause major accidents; in flammable and explosive environments, fire or explosion may result; portable wearable human body equipment is extremely inconvenient and extremely high in cost if a rechargeable battery is not used. However, these problems can be solved well if wireless energy transmission technology is used.
As one of the hottest emerging technologies in the 21 st century, the magnetic coupling resonance Wireless energy Transfer (WPT) technology has the advantages of high transmission efficiency, strong safety, small influence of transmission distance and direction, and the like, and has a very wide application prospect: at present, the magnetic coupling resonance WPT technology is used as an enterprise employee card, and is used as an industrial environment which is closely related to the life of each person or is vigorously developed in various countries for electric energy transmission of large-scale equipment in space, so that the magnetic coupling resonance WPT technology has great value.
However, the technology still has high potential practical value, so that the technology becomes a research hotspot of various researchers in recent years, with the continuous development of the litz wire winding alternating current resistance technology, an optimization scheme for the litz wire winding design can be continuously updated in an iterative manner, which is beneficial to reducing energy loss in a WPT system and greatly reducing the transportation cost of energy, but the current mainstream mode is to obtain the alternating current resistance of the litz wire winding by using a manual calculation mode through a research method for simplifying a calculation formula, and the problems of long time consumption, complex calculation, inaccuracy and the like of an analytical calculation method are solved.
Disclosure of Invention
The invention aims to provide a time-saving method for automatically calculating the litz wire winding alternating current resistance aiming at the defect that the analysis in the prior art needs a large amount of manual calculation, which is used for reducing the time required by calculation; because the automatic operation can reduce the errors caused by manual work, the calculation accuracy of the alternating current resistance is improved by continuously perfecting and training the litz wire winding neural network model, the litz wire winding is automatically calculated through the neural network model, and the optimal resistance value can be quickly and accurately calculated.
The technical scheme adopted by the invention is an intelligent design optimization method for litz wire winding, which is divided into two parts: the first part is to establish a litz wire winding simulation model based on hfss software, and the second part is to optimize the litz wire winding simulation model based on a neural network model.
S1, establishing a litz wire winding simulation model based on hfss software;
step 1.1: a plane litz wire winding simulation model is established through hfss simulation software to obtain alternating current resistance values under different parameters, various data under different frequencies, different materials, different sizes and different litz wire winding radiuses are obtained, and neural network training and testing are performed on the obtained various data. The winding types of the plane litz wire winding simulation model are circular wires, rectangular wires and square wires.
Step 1.2: in hfss simulation software, the difference between the simulation result of the built plane litz wire winding with different types and the actual result between the actual litz wire winding is analyzed, the difference between the errors of the wires with different types is confirmed to be small through comparison, and square wires are selected as the winding types in order to save simulation time. And then, alternating current resistance values under different parameters are obtained through hfss simulation, and a large amount of data under different frequencies, different materials (different materials cause different conductivities), different sizes and different litz wire winding radiuses are obtained to train and test the neural network.
S2, optimizing a litz wire winding simulation model based on a neural network model;
and (3) rapidly obtaining the porosity and the alternating current resistance of the litz wire winding by the neural network based on the particle swarm optimization algorithm, wherein the porosity is the ratio of the litz wire diameter to the litz wire spacing. The method comprises the following specific steps:
step 2.1: and creating a training set and a testing set for training and testing in matlab by using a simulation result obtained by the litz wire-wound simulation model established in hfss in the last step. In the neural network, the litz line alternating current resistance and the porosity are used as output values, the geometric size and the frequency of the litz line are used as input values, and in each step of a particle group optimization algorithm in the training process, each litz line size is searched around the minimum point once found by the particle group optimization algorithm and the minimum point found by the whole litz line size group. And (4) taking the minimum point which is searched by the litz line size group as the minimum point of the alternating current resistance function of the litz line through iteration. Each litz line size in the algorithm represents one possible solution of the litz line ac resistance function.
Step 2.2: the size of each litz line size (e.g. litz line radius) and the frequency of each litz line are updated in each iteration under the influence of the other litz line sizes (other sizes than the litz line radius), the updating of their frequency and size differences following equations (1-1) and (1-2).
WhereinFor a single litz line size extreme,for a plurality of litz line size extremes,for the previous single litz line size,current single litz line size. Equation (1-1) consists of three terms, in the first of which w is a weight that adjusts the current frequency of the litz lineDetermines to what extent the litz line should maintain its previous frequencyThe second term is the size-poor cognitive component, factorAdjusting the size of the litz line dimension, wherein rand refers to a random number in the middle of 0 to 1. The last term represents the litz line size adjustment, which is scaled by a factorThe adjustment, defines a size difference relative to the set of best solutions. Parameter(s)Andhow much weight should be given to each of the search results that perfect the litz line itself and the search results that identify the size group of the litz line.
Step 2.3: single litz line size limitAnd a plurality of litz line size extremesIt is updated with the adaptation value of the litz line size being updated in each new iteration. The two are the location of the optimal solution explored by a single litz line size and the size of the optimal solution explored by all the litz line sizes in the plurality of litz line size groups, respectively.
The PSO algorithm requires only these several parameters to operate: the value of the limiting factor (w,and) The number of litz line sizes, the maximum number of iterations, and a fitness litz line ac resistance function. The function of the fitness litz line AC resistance function is to evaluate the performance of the litz line AC resistance in each iteration and select the optimal porosity scheme which enables the litz line AC resistance under the current litz line condition to be minimum.
Compared with the prior art, the litz line winding alternating current resistance and the porosity are obtained through the neural network and the hfss, and the litz line winding alternating current resistance and the porosity are obtained through manual calculation in the prior art, and both the litz line winding alternating current resistance and the porosity are concentrated on simplified formulas, so that the calculation amount and the material loss can be greatly reduced. For the calculation of the alternating current resistance of the litz wire winding, the alternating current resistance of the litz wire winding is obtained by using an artificial calculation mode through a research method of simplifying a calculation formula, and the method has the problems of long time consumption, complex calculation, inaccuracy and the like in analytic calculation. In the method, model construction is carried out on the litz wire winding by using hfss (high-frequency electromagnetic simulation software) simulation software to simulate to obtain the alternating current resistance values of the litz wire winding under different materials, different turns, different litz wire winding radiuses and different frequencies, so that training data and test data used by the litz wire winding neural network model are completed. For the simulation of the litz line winding model, a square line meshing mode is adopted to replace the meshing mode of litz line winding circular lines in the meshing involved with the mode driving solving and the frequency sweeping mode, and the manual meshing is adopted, so that the solving step and time are simplified, and the accuracy is ensured. And then automatically training and testing by using a litz wire winding neural network model by using a particle swarm optimization algorithm to perfect the network model, receiving data obtained by simulation of the litz wire winding model transmitted by hfss through the trained network model to improve the accuracy of the litz wire winding neural network model, and finally obtaining the required litz wire winding alternating current resistance quickly and accurately by using the litz wire winding neural network model. The method can be used for quickly and accurately obtaining the litz wire winding alternating current resistance under different requirements, reduces the calculated amount and the working time, and greatly improves the efficiency.
Drawings
FIG. 1 shows the litz wire winding model to be built in hfss.
Fig. 2 is a flow chart of a function extremum optimizing algorithm of the PSO algorithm.
FIG. 3 is a combined structure diagram of hfss and litz wire winding neural network model.
Detailed Description
Fig. 1 shows the planar litz wire winding ac resistance to be established in hfss. Firstly, establishing plane litz line winding alternating current resistance in hfss, dividing a circular line into grids in a grid division mode of a square line through the comparison of the square line when a model is established, reducing simulation time, wherein the litz line winding alternating current resistance is formed by winding different materials and different turns, and training data and test data are obtained through an hfss frequency scanning mode and a variable scanning mode to train a neural network. The obtained data is made into a training set and a testing set through files for use.
The function extremum optimizing algorithm of the PSO algorithm of fig. 2 is a flow chart. And then, a neural network is built in the matlab, a particle swarm optimization algorithm is realized in the matlab, and a function extremum optimization algorithm based on a PSO algorithm is adopted.
The design method for realizing the neural network of the two-layer perceptron in matlab is as follows.
PSO Algorithm parameter settings
Individual code length according to formula (1)-3) calculation, population size 30, evolution number 100. Particle swarm algorithm parametersAndall are 1.49445.
length=inpnt n *hidden n +hidden n *output n +hidden n +output n (1-3)
Wherein, length is the individual coding length, input is the input grain, output is the output value, and hidden is the number of hidden layers.
Litz line size initialization
litz line size initialization means that a random value is given to the size and frequency of each litz line in all the litz line sizes, so that the litz line size adaptability value is calculated according to the adaptability litz line alternating current resistance function. According to the method, the weight of the network is updated through the node numbers of the input layer, the output layer and the hidden layer, and the network is trained.
3. Finding the initial extremum
And according to the initial randomly assigned litz line size fitness, searching single and multiple extreme litz line sizes in the group.
4. Iterative optimization
The litz line frequency and size are updated according to equations (1-1) and (1-2). And calculating the fitness value of the litz line according to the formula (1-1), and determining a new single extreme value and a plurality of extreme values.
FIG. 3 is a block diagram of the hfss and litz wire-wound neural network model, which is combined through matlab codes, so that the data does not need to be manually updated each time, and the litz wire-wound neural network model is automatically called from hfss.
Examples
A litz wire winding intelligent design optimization method is divided into two parts: the first part is to establish a litz wire winding simulation model based on hfss software, and the second part is to optimize the litz wire winding simulation model based on a neural network model.
S1, establishing a litz wire winding simulation model based on hfss software;
step 1.1: because the alternating current resistance values of the litz wire windings are different under different frequencies, the alternating current resistance values of the litz wire windings are different when the number of turns, the radius and the like of the litz wire windings are different, calculation needs to be repeatedly carried out through a formula, the method is very complex and time-consuming, meanwhile, the planar litz wire windings are wound by metal wires into induction litz wire windings, the litz wire windings are on the same plane and are sequentially wound from inside to outside, and different materials can also influence the alternating current resistance values of the litz wire windings under different environments, so that the required time is increased; alternating current resistance values under different parameters are obtained by establishing a planar litz wire winding through hfss simulation software, and a large amount of data under different frequencies, different materials (different materials cause different conductivities), different sizes and different litz wire winding radiuses are obtained to train and test the neural network.
Step 1.2: in hfss, simulation results and actual result differences between different kinds of planar litz wire windings (circular wire, rectangular wire and square wire) and actual litz wire windings are compared, and the results are as follows:
the result shows that the error difference between different types of lines is small, but the grid division between different types is different, the simulation time is also different, and the simulation time can be greatly reduced for the uniform grid division of the rectangular lines, so that in order to reduce the simulation time and improve the calculation efficiency, the circular lines are replaced by the uniform grids of the square lines during simulation, and the simulation data can be quickly called when the neural network calls the simulation data.
Establishing and perfecting a second part neural network model
The neural network used in the patent uses a Particle Swarm Optimization (PSO), and optimizes data through the PSO so that the PSO can quickly obtain a proper porosity, i.e., a ratio of a wire diameter to a wire spacing. PSO is a simple algorithm that searches for an optimal solution in the solution space. It, unlike other optimization algorithms, requires only the target function and does not rely on any differential form of the gradient or target. It also requires very few hyper-parameters.
The PSO is best suited to work to find the maximum or minimum of a function defined on a multidimensional vector space, and is optimized in a manner similar to a bird flock finding food, starting with random points, i.e. particles, on a plane, letting them find the minimum in random directions. The method comprises the following specific steps:
step 2.1: at each step, each litz line size should be searched around the minimum point it has found and the minimum point found for the entire litz line size cluster. After a certain iteration, the minimum point which is once explored by the litz line size group is taken as the minimum point of the litz line alternating current resistance function. Thus, each litz line size in the algorithm represents one possible solution of the litz line ac resistance function.
Step 2.2: the size of each litz line size (e.g. litz line radius) and the frequency of each litz line are updated in each iteration under the influence of the other litz line sizes (other sizes than the litz line radius), the updating of their frequency and size difference following equations (1-1) and (1-2).
Equation (1-1) consists of three terms, in the first of which w is a weight that adjusts the current frequency of the litz line, determining how much the litz line should maintain its previous frequency. The second term is the cognitive component of the size difference, the factor c1 regulates the size of the litz line size, where rand refers to a random number in the middle of 0 to 1. The last term represents the litz line size adjustment, which is adjusted by a factor c2, defining the size difference relative to the set of best solutions. The parameters c1 and c2 control how much weight should be given to each of the search results that refine the litz line itself and the search results that identify the litz line size group.
Step 2.3: the single litz line size extreme xPbesth and the multiple litz line size extremes xgtest will be updated with the fitness value of the updated litz line size at each new iteration. The two are the location of the optimal solution explored by a single litz line size and the size of the optimal solution explored by all the litz line sizes in the plurality of litz line size groups, respectively.
The PSO algorithm requires only these several parameters to operate: limiting factor values (w, c1 and c 2), number of litz line sizes, maximum number of iterations, and a fitness litz line ac resistance function. The fitness litz line alternating current resistance function has the function of evaluating the performance of litz line alternating current resistance in each iteration and selecting the best solution.
Step 2.4: this patent is through simplifying the simulation model of plane litz line winding alternating current resistance, carries out data packing integration again and sends to neural network in, obtains the optimal plane litz line winding size and porosity etc. that correspond to under this external condition through neural network, reduction calculated amount and material loss that can be very big.
Claims (5)
1. A litz line winding intelligent design optimization method is characterized by comprising the following steps,
s1, establishing a litz wire winding simulation model based on hfss software;
step 1.1: establishing a plane litz wire winding simulation model through hfss simulation software to obtain alternating current resistance values under different parameters, obtaining various data under different frequencies, different materials, different sizes and different litz wire winding radiuses, and performing neural network training and testing on the obtained various data; the winding types of the plane litz wire winding simulation model are circular wires, rectangular wires and square wires;
step 1.2: in hfss simulation software, analyzing the difference between the simulation result of the built plane litz wire windings of different types and the actual result between the actual litz wire windings, comparing to confirm that the difference between the errors of the wires of different types is small, and selecting a square wire as the winding type; alternating current resistance values under different parameters are obtained through hfss simulation, and data under different frequencies, different materials, different sizes and different litz wire winding radiuses are obtained to conduct neural network training and testing;
s2, optimizing a litz wire winding simulation model based on a neural network model;
the method comprises the steps that a neural network based on a particle swarm optimization algorithm quickly obtains the porosity and the alternating current resistance of a litz wire winding, wherein the porosity is the ratio of the litz wire diameter to the litz wire distance; the method comprises the following specific steps:
step 2.1: utilizing the litz wire-winding simulation model established in the hfss simulation in the last step to establish a training set and a testing set for training and testing in matlab according to the obtained simulation result; in the neural network, the litz line alternating current resistance and the porosity are used as output values, the geometric size and the frequency of the litz line are used as input values, and in each step of the particle group optimization algorithm in the training process, the size of each litz line is searched around the found minimum point and the minimum point found in the whole litz line size group; after iteration, taking the minimum point once explored by the litz line size group as the minimum point of the litz line alternating current resistance function; each litz line size represents one possible solution of the litz line ac resistance function;
step 2.2: the size of each litz line size and the frequency of each litz line are updated in each iteration under the influence of other litz line sizes, and the updating of the frequency and size difference follows the formulas (1-1) and (1-2);
whereinFor a single litz line size extreme,for a plurality of litz line size extremes,for the previous single litz line size,current single litz line size; equation (1-1) consists of three terms, in the first of which w is the weight that adjusts the current frequency of the litz lineDetermine to what extent the litz line should maintain its previous frequencyThe second term is the size-poor cognitive component, factorAdjusting the size of the litz line size; rand means a random number in the middle of 0 to 1; the last term represents the litz line size adjustment, by a factorAdjusting, defining a size difference relative to the set of optimal solutions; parameter(s)Andcontrolling how much weight should be obtained for perfecting the search result of the litz line and the search result for identifying the size group of the litz line;
step 2.3: single litz line size extremeAnd a plurality of litz line size extremesWill be updated with the updated fitness value of the litz line size in each new iteration; both are the location of the optimal solution explored by a single litz line size, and the size of the optimal solution explored by all the litz line sizes in the multiple litz line size clusters, respectively.
2. The litz wire winding intelligent design optimization method as claimed in claim 1, characterized in that in order to facilitate accurate calculation of electromagnetic field distribution inside the conducting wire in the litz wire winding simulation model, a square wire meshing manner is used instead of a circular wire meshing manner of the litz wire winding.
3. The litz wire winding intelligent design optimization method as claimed in claim 1, wherein the litz wire winding model built by using hfss is self-adjustable, a certain frequency range and litz wire winding size variables are set in hfss, and different environments are simulated by changing the variable values and the frequency range to meet different requirements.
4. The litz wire-wound intelligent design optimization method as claimed in claim 1, wherein the neural network uses PSO (particle swarm optimization).
5. The method as claimed in claim 1, wherein the neural network autonomously selects whether to change the number of neurons to reach the set convergence value, and calls hfss simulation results at any time through codes to complete the litz wire winding neural network model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211161074.4A CN115455837A (en) | 2022-09-22 | 2022-09-22 | Intelligent design optimization method for litz line winding |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211161074.4A CN115455837A (en) | 2022-09-22 | 2022-09-22 | Intelligent design optimization method for litz line winding |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115455837A true CN115455837A (en) | 2022-12-09 |
Family
ID=84307633
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211161074.4A Pending CN115455837A (en) | 2022-09-22 | 2022-09-22 | Intelligent design optimization method for litz line winding |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115455837A (en) |
-
2022
- 2022-09-22 CN CN202211161074.4A patent/CN115455837A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110059377A (en) | A kind of fuel battery service life prediction technique based on depth convolutional neural networks | |
CN107482797B (en) | Annular coil electromagnetic radiation prediction method based on resonant wireless power transmission | |
CN110264012A (en) | Renewable energy power combination prediction technique and system based on empirical mode decomposition | |
CN110212529A (en) | A kind of power distribution network low-voltage platform area method for early warning based on support vector machines | |
CN106779065A (en) | The maximum power point tracing method and system of a kind of photovoltaic array | |
CN107871024B (en) | Electromagnetic optimization method and device for high-temperature superconductive annular energy storage magnet | |
CN103077316A (en) | Peak-clipping and valley-filling optimizing method of load curve | |
CN114744310B (en) | Power lithium battery two-step charging method based on SOC self-adaptive grading | |
CN115455837A (en) | Intelligent design optimization method for litz line winding | |
CN111178528B (en) | Elite genetic algorithm improvement method applied to wireless charging system | |
CN111009973B (en) | Resonance coil for resisting deviation in wireless power transmission | |
CN103237312B (en) | A kind of wireless sensor network node coverage optimization method | |
CN106021706B (en) | High performance induction motor light weight method based on the collaboration optimization of population multiple physical field | |
Du et al. | Optimization of magnetically coupled resonant wireless power transfer based on improved whale optimization algorithm | |
CN115310353B (en) | Power transformer design method based on rapid multi-objective optimization | |
CN114784324B (en) | Fuel cell system control method and device, electronic equipment and storage medium | |
CN108053250B (en) | Genetic algorithm-based transformer price optimization method and device | |
CN112487723B (en) | Multi-parameter-based optimized design method for reclaimed water heat source pump system | |
CN108770001B (en) | Wireless chargeable sensor network optimization method based on memetic algorithm | |
CN109086952A (en) | It is a kind of based on genetic algorithm-neural network heat load prediction method | |
CN112434461B (en) | Automatic coupling coil optimization design method for wireless power transmission | |
CN110737998B (en) | Grading ring optimization design method based on finite element and deep belief network | |
Li et al. | A base station location planning method based on artificial bee colony algorithm | |
Wu et al. | Precise Modeling of the Self-Inductance of Circular Coils with Deep Neural Networks | |
Jabri et al. | Misalignment controller in wireless battery charger for electric vehicle based on MPPT method and metaheuristic algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |