CN115455505A - Active construction method of full-freedom permanent magnet - Google Patents
Active construction method of full-freedom permanent magnet Download PDFInfo
- Publication number
- CN115455505A CN115455505A CN202211031315.3A CN202211031315A CN115455505A CN 115455505 A CN115455505 A CN 115455505A CN 202211031315 A CN202211031315 A CN 202211031315A CN 115455505 A CN115455505 A CN 115455505A
- Authority
- CN
- China
- Prior art keywords
- magnetic
- magnet
- state
- magnetic block
- magnetic field
- 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.)
- Granted
Links
- 238000010276 construction Methods 0.000 title claims abstract description 19
- 230000009471 action Effects 0.000 claims abstract description 47
- 238000011156 evaluation Methods 0.000 claims abstract description 37
- 230000003993 interaction Effects 0.000 claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 239000000463 material Substances 0.000 claims abstract description 9
- 238000004519 manufacturing process Methods 0.000 claims abstract description 7
- 238000000034 method Methods 0.000 claims description 27
- 239000013598 vector Substances 0.000 claims description 23
- 230000002787 reinforcement Effects 0.000 claims description 15
- 230000004927 fusion Effects 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 8
- 230000006872 improvement Effects 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 8
- 230000005347 demagnetization Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 238000003384 imaging method Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 230000001186 cumulative effect Effects 0.000 claims description 4
- 239000000696 magnetic material Substances 0.000 claims description 2
- 230000007704 transition Effects 0.000 claims 1
- 238000005457 optimization Methods 0.000 description 7
- 238000002595 magnetic resonance imaging Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000002068 genetic effect Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 229910052734 helium Inorganic materials 0.000 description 1
- 239000001307 helium Substances 0.000 description 1
- SWQJXJOGLNCZEY-UHFFFAOYSA-N helium atom Chemical compound [He] SWQJXJOGLNCZEY-UHFFFAOYSA-N 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 229910001172 neodymium magnet Inorganic materials 0.000 description 1
- 229910052761 rare earth metal Inorganic materials 0.000 description 1
- 150000002910 rare earth metals Chemical class 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 210000004872 soft tissue Anatomy 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- 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
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
本发明涉及一种全自由度永磁体主动构建方法，包括以下步骤：步骤1、选取永磁材料并制作磁块；步骤2、构建全自由度磁体几何模型；步骤3、确定由少量磁块组成的磁体初始状态；步骤4、构建磁场评价环境；步骤5、构建基于深度神经网络的磁块动作决策网络；步骤6、进行磁块动作决策网络与磁场评价环境的交互，将交互产生的经验保存到回放单元；步骤7、依据收敛的永磁体模型制造永磁体。本发明能够在同等重量条件下提升永磁体的磁场场强并优化磁场的均匀性。
The invention relates to an active construction method of a full-degree-of-freedom permanent magnet, comprising the following steps: step 1, selecting permanent magnet materials and making magnetic blocks; step 2, constructing a full-degree-of-freedom magnet geometric model; step 3, determining that it is composed of a small number of magnetic blocks The initial state of the magnet; step 4, construct the magnetic field evaluation environment; step 5, construct the magnetic block action decision network based on the deep neural network; step 6, carry out the interaction between the magnetic block action decision network and the magnetic field evaluation environment, and save the experience generated by the interaction Go to the playback unit; step 7, manufacture the permanent magnet according to the converged permanent magnet model. The invention can increase the magnetic field intensity of the permanent magnet and optimize the uniformity of the magnetic field under the same weight condition.
Description
技术领域technical field
本发明属于信息与通信工程技术领域,涉及一种永磁体主动构建方法，尤其是一种全自由度永磁体主动构建方法。The invention belongs to the technical field of information and communication engineering, and relates to an active construction method of a permanent magnet, in particular to an active construction method of a full-degree-of-freedom permanent magnet.
背景技术Background technique
磁共振成像(Magnetic resonance imaging，MRI)具有零辐射暴露和良好的软组织成像对比度等优点，主磁体作为磁共振成像仪最基本的构件，其性能将直接影响磁共振图像的质量。Magnetic resonance imaging (MRI) has the advantages of zero radiation exposure and good soft tissue imaging contrast. The main magnet is the most basic component of an MRI apparatus, and its performance will directly affect the quality of MRI images.
现有的主磁体主要分为超导电磁型和永磁型两种。超导电磁型主磁体能够实现场强高、均匀性好的磁场，从而获得相较于传统永磁型磁体质量更好的磁共振成像，是目前主流的市场选择，但是超导磁体造价高，且需要定期补充稀缺的液氦，维护费用较高，导致磁共振检测成本高昂，限制了磁共振设备的吞吐量。相比而言，永磁体一般采用稀土永磁材料制成，如我国储量丰富的钕铁硼，因此造价低廉，易于普及。此外永磁体还具有低耗能、维护费用低等优点。但受限于较差的均匀度，基于永磁体的磁共振通常成像质量较差，因此研究高均匀性的永磁体能够降低单台磁共振部署成本，提升社会面磁共振部署总数，从而提升磁共振检测的总体吞吐量，对日益严重的老龄化社会具有重要意义。Existing main magnets are mainly divided into superconducting electromagnetic type and permanent magnet type. The superconducting electromagnetic main magnet can achieve high field strength and good uniformity of the magnetic field, so as to obtain magnetic resonance imaging with better quality than traditional permanent magnets. It is the current mainstream market choice, but the cost of superconducting magnets is high. Moreover, the scarce liquid helium needs to be replenished regularly, and the maintenance cost is high, which leads to high cost of magnetic resonance detection and limits the throughput of magnetic resonance equipment. In contrast, permanent magnets are generally made of rare earth permanent magnet materials, such as NdFeB, which is abundant in my country, so they are cheap and easy to popularize. In addition, permanent magnets also have the advantages of low energy consumption and low maintenance costs. However, limited by poor uniformity, permanent magnet-based magnetic resonance usually has poor imaging quality. Therefore, research on permanent magnets with high uniformity can reduce the cost of a single magnetic resonance deployment, increase the total number of magnetic resonance deployments in the society, and thus improve the magnetic resonance imaging quality. The overall throughput of resonance detection is of great significance for an increasingly aging society.
永磁体一般由多个永磁材料制成的磁块按照特殊设计的空间分布构成，磁块的空间分布决定了永磁体的均匀性，因此研究者一般通过对磁块空间排布优化的研究来提升永磁体的均匀性。目前研究者对永磁体磁块排布的优化通常基于遗传算法，每个磁块的角度都固定为Halbach型磁体的角度，因此算法仅能确定少量的参数，如某个位置磁块的型号。这类优化方法存在两个问题，首先遗传算法不涉及对磁场特征的提取，通常为磁块组合的随机试错，难以建模磁块间复杂的剩磁退磁效应以及每个磁块对总体磁场的影响；此外磁块的角度固定，限制了主磁体的理论性能上限。因此，虽然基于遗传算法的磁块排布优化已经在主磁体均匀性上取得了显著的突破，但仍有很大的提升空间。Permanent magnets are generally composed of multiple magnetic blocks made of permanent magnetic materials according to a specially designed spatial distribution. The spatial distribution of the magnetic blocks determines the uniformity of the permanent magnets. Therefore, researchers generally study the optimization of the spatial arrangement of the magnetic blocks. Improve the uniformity of permanent magnets. At present, researchers usually optimize the arrangement of permanent magnet magnets based on genetic algorithm, and the angle of each magnet is fixed to the angle of Halbach-type magnets. Therefore, the algorithm can only determine a small number of parameters, such as the model of a magnet at a certain position. There are two problems in this type of optimization method. First, the genetic algorithm does not involve the extraction of magnetic field characteristics. It is usually a random trial and error combination of magnetic blocks. It is difficult to model the complex remanence demagnetization effect between magnetic blocks and the effect of each magnetic block on the overall magnetic field. In addition, the angle of the magnetic block is fixed, which limits the upper limit of the theoretical performance of the main magnet. Therefore, although the optimization of the magnetic block arrangement based on the genetic algorithm has made a significant breakthrough in the uniformity of the main magnet, there is still a lot of room for improvement.
经检索，未发现与本发明相同或相似的现有技术的专利文献。After searching, no patent documents of the prior art identical or similar to the present invention are found.
发明内容Contents of the invention
本发明的目的在于克服现有技术的不足，提出一种全自由度永磁体主动构建方法,能够在同等重量条件下提升永磁体的磁场场强并优化磁场的均匀性。The purpose of the present invention is to overcome the deficiencies of the prior art, and propose an active construction method for a full-degree-of-freedom permanent magnet, which can increase the magnetic field strength of the permanent magnet and optimize the uniformity of the magnetic field under the same weight condition.
本发明解决其现实问题是采取以下技术方案实现的：The present invention solves its practical problems and is realized by taking the following technical solutions:
一种全自由度永磁体主动构建方法，包括以下步骤：An active construction method for a full-degree-of-freedom permanent magnet, comprising the following steps:
步骤1、选取永磁材料并制作磁块；
步骤2、构建全自由度磁体几何模型；Step 2. Construct a full-degree-of-freedom magnet geometric model;
步骤3、基于步骤2构建的全自由度磁体几何模型，确定由少量磁块组成的磁体初始状态，设定初始磁块为原始Halbach角度；Step 3. Based on the full-degree-of-freedom magnet geometric model constructed in step 2, determine the initial state of the magnet composed of a small number of magnetic blocks, and set the initial magnetic block to the original Halbach angle;
步骤4、在强化学习的框架下，构建磁场评价环境；Step 4. Under the framework of reinforcement learning, construct a magnetic field evaluation environment;
步骤5、构建基于深度神经网络的磁块动作决策网络，对磁体状态中的磁场状态及其对应磁块排布状态进行不同方式的特征提取与融合，最终实现磁体状态到Q值的精确映射；Step 5. Construct a decision-making network for magnetic block action based on a deep neural network, perform feature extraction and fusion in different ways on the magnetic field state in the magnet state and the corresponding magnetic block arrangement state, and finally realize the accurate mapping from the magnet state to the Q value;
步骤6、基于步骤4构建的磁体评价环境和步骤5构建的基于深度神经网络的磁块动作决策网络，依据步骤3预定的初始状态与设定的重量约束条件进行磁块动作决策网络与磁场评价环境的交互，将交互产生的经验保存到回放单元；Step 6. Based on the magnet evaluation environment built in step 4 and the magnet block action decision network based on the deep neural network built in step 5, the magnet block action decision network and magnetic field evaluation are performed according to the initial state predetermined in step 3 and the set weight constraints The interaction of the environment saves the experience generated by the interaction to the playback unit;
步骤7、将步骤6所得经验输入到步骤5所述磁块动作决策网络中，对磁块动作决策网络进行优化，依据收敛的永磁体模型制造永磁体；Step 7, input the experience obtained in step 6 into the magnetic block action decision network described in step 5, optimize the magnetic block action decision network, and manufacture permanent magnets according to the converged permanent magnet model;
而且，所述步骤1的具体方法为：And, the concrete method of described
根据磁能积、剩磁温度系数与可加工性选取永磁材料，并根据需求制作不同几何尺寸的磁块；Select permanent magnet materials according to magnetic energy product, remanence temperature coefficient and machinability, and make magnets of different geometric sizes according to requirements;
而且，所述步骤2的具体方法为：And, the concrete method of described step 2 is:
构建全自由度磁体几何模型，在该磁体几何模型中，磁块的方向自由、磁块在模型中的候选位置自由、磁体参数自由。其中磁体参数包括行、列和层数，“行”代表磁体轴向的“磁块环”、“列”代表与磁体轴向平行的“磁块列”、相同孔径的所有相邻列共同组成“层”。A full-degree-of-freedom magnet geometric model is constructed. In the magnet geometric model, the direction of the magnet block is free, the candidate position of the magnet block in the model is free, and the magnet parameters are free. Among them, the magnet parameters include row, column and layer number, "row" represents the "magnetic block ring" in the axial direction of the magnet, "column" represents the "magnetic block column" parallel to the magnet axial direction, and all adjacent columns with the same aperture are jointly composed "Floor".
而且，所述步骤3的具体方法为：And, the concrete method of described step 3 is:
在上述磁体几何模型中，首先确定由少量磁块组成的磁体初始状态，磁体的初始状态包括在磁体几何模型中若干层的若干中间行相对均匀分布的若干磁块，设定初始磁块为原始Halbach角度。In the above geometric model of the magnet, the initial state of the magnet composed of a small number of magnetic blocks is firstly determined. The initial state of the magnet includes a number of relatively evenly distributed magnetic blocks in several middle rows of several layers in the geometric model of the magnet. The initial magnetic block is set as the original Halbach angle.
而且，所述步骤4的具体方法为：And, the concrete method of described step 4 is:
在强化学习的框架下，构建基于有限元分析的包含有剩磁退磁效应建模的磁场评价环境，该磁场评价环境用于实现磁体状态的转移以及磁块动作决策前后成像区域磁场均值与均匀度改善程度的评估；Under the framework of reinforcement learning, a magnetic field evaluation environment including remanence demagnetization effect modeling based on finite element analysis is constructed. This magnetic field evaluation environment is used to realize the transfer of the magnet state and the average and uniformity of the magnetic field in the imaging area before and after the decision-making of the magnetic block. assessment of improvement;
所述均匀度改善程度称为奖励R，可用的评价指标包括平均磁场强度Bmean和磁场均匀度Bhomogeneous。The improvement degree of uniformity is called reward R, and available evaluation indexes include average magnetic field strength B mean and magnetic field uniformity B homogeneous .
而且，所述步骤5的具体步骤包括：And, the specific steps of described step 5 include:
(1)对于磁场状态，构建深度神经网络对其进行特征提取，得到磁场特征；(1) For the state of the magnetic field, construct a deep neural network for feature extraction to obtain magnetic field features;
(2)对于磁场状态对应的磁块排布状态，用磁块状态向量编码磁块状态，一个磁块状态子向量长度为n+1，表示磁块状态子向量中用1个标量来编码磁块位置是否为空，用n个标量来编码磁块的角度；(2) For the magnetic block arrangement state corresponding to the magnetic field state, the magnetic block state is encoded by the magnetic block state vector, and the length of a magnetic block state subvector is n+1, which means that one scalar is used to encode the magnetic block state subvector. Whether the block position is empty, use n scalars to encode the angle of the magnetic block;
(3)将磁块状态向量输入深度神经网络得到磁块排布特征；(3) Input the magnetic block state vector into the deep neural network to obtain the magnetic block arrangement feature;
(4)将磁场特征与磁块排布特征输入到深度神经网络的特征融合模块进行特征的交互与融合；(4) Input the magnetic field feature and the magnetic block arrangement feature into the feature fusion module of the deep neural network to perform feature interaction and fusion;
(5)在深度神经网络最后的分类层得到磁体状态到Q值的映射，其中Q值在深度强化学习中代表奖励值的累积折扣期望。(5) In the final classification layer of the deep neural network, the mapping of the magnet state to the Q value is obtained, where the Q value represents the cumulative discounted expectation of the reward value in deep reinforcement learning.
而且，所述步骤6的具体步骤包括：And, the specific steps of described step 6 include:
(1)在步骤3预定的磁体初始状态下，以磁体重量W为终止条件进行步骤5所述磁块动作决策网络与步骤4所述磁场评价环境的交互，在达到重量上限后返回步骤3所述磁体初始状态并重复交互过程；(1) In the initial state of the magnet predetermined in step 3, the magnet weight W is used as the termination condition to carry out the interaction between the magnet block action decision network described in step 5 and the magnetic field evaluation environment described in step 4, and return to step 3 after reaching the upper limit of weight Describe the initial state of the magnet and repeat the interaction process;
(2)将交互过程中产生的磁块动作决策网络与磁场评价环境的互动经验保存到回放单元，每一条经验包括当前磁体数据状态、磁块动作、奖励以及下一个磁体数据状态；(2) Save the interaction experience between the magnetic block action decision network and the magnetic field evaluation environment generated during the interaction process to the playback unit, each experience includes the current magnet data state, magnetic block action, reward and next magnet data state;
而且，所述步骤7的具体方法为：And, the concrete method of described step 7 is:
从步骤6所述回放单元中取出经验对步骤5所述磁块动作决策网络进行优化，当磁块动作决策网络收敛时即视为构建出了最优性能的永磁体模型，根据主动构建的全自由度永磁体模型即可制造主磁体。Take experience from the playback unit described in step 6 to optimize the magnetic block action decision-making network described in step 5. When the magnetic block action decision-making network converges, it is considered to have constructed a permanent magnet model with optimal performance. The main magnet can be manufactured from the permanent magnet model with degrees of freedom.
本发明的优点和有益效果：Advantages and beneficial effects of the present invention:
1、本发明针对当前磁块排布优化算法以及磁体模型存在的问题，提出一种全自由度永磁体主动构建方法，在深度强化学习的算法框架下，构建全自由度磁体几何模型以破除传统Halbach型永磁体的理论性能上限。并提出三维磁块动作决策网络以提取磁场及磁块排布的多模态特征，从而在全自由度的磁体几何模型中依据多模态的深层次融合特征，在磁体重量相同时主动构建出相较于Halbach型永磁体磁场强度更高、磁场均匀性更好的永磁体。1. Aiming at the problems existing in the current optimization algorithm of magnetic block arrangement and magnet model, the present invention proposes a full-degree-of-freedom permanent magnet active construction method. Under the algorithm framework of deep reinforcement learning, a full-degree-of-freedom magnet geometric model is constructed to break the traditional The theoretical performance upper limit of Halbach-type permanent magnets. And a three-dimensional magnetic block action decision network is proposed to extract the multi-modal features of the magnetic field and the arrangement of the magnetic blocks, so that in the full-degree-of-freedom magnet geometric model, according to the multi-modal deep-level fusion features, when the weight of the magnet is the same, it is actively constructed. Compared with the Halbach type permanent magnet, it is a permanent magnet with higher magnetic field strength and better magnetic field uniformity.
2、本发明提出一种全自由度永磁体主动构建方法，根据最大磁能积、剩磁温度系数与可加工性等需求选取永磁材料并制作不同几何尺寸的磁块，设计磁块方向自由、位置自由，磁体几何模型参数自由的全自由度磁体几何模型，在强化学习的框架下，构建基于有限元分析方法的具有剩磁退磁效应建模的磁场评价环境，同时构建基于深度神经网络的磁块动作决策网络，以磁体重量为约束，在磁块动作决策网络与磁场评价环境的持续互动中，使决策网络具备从磁体状态到奖励值的精确映射能力。上述磁场评价环境和磁块动作决策网络能够使强化学习智能体在固定磁体重量上限时，主动探索并优化磁块排布，从而使目标成像区域的磁场强度更高、均匀性更好。2. The present invention proposes a full-degree-of-freedom permanent magnet active construction method. According to the requirements of maximum magnetic energy product, remanence temperature coefficient and machinability, permanent magnet materials are selected and magnetic blocks of different geometric sizes are made. The direction of the magnetic block is designed to be free, The full-degree-of-freedom magnet geometric model with free position and free magnet geometric model parameters, under the framework of reinforcement learning, constructs a magnetic field evaluation environment with remanence and demagnetization effect modeling based on finite element analysis method, and constructs a magnetic field evaluation environment based on deep neural network at the same time. The block action decision network is constrained by the weight of the magnet. In the continuous interaction between the magnet block action decision network and the magnetic field evaluation environment, the decision network has the ability to accurately map from the magnet state to the reward value. The above-mentioned magnetic field evaluation environment and magnetic block action decision-making network can enable the reinforcement learning agent to actively explore and optimize the arrangement of magnetic blocks when the upper limit of the weight of the magnet is fixed, so that the magnetic field intensity of the target imaging area is higher and the uniformity is better.
3、本发明提出了一种基于深度神经网络的全自由度永磁体主动构建方法，尽可能减少了对磁体几何模型的限制，同时运用深度强化学习算法对磁块排布进行主动探索，能够在轻重量约束的条件下构建出场强高，均匀性好的永磁体模型。3. The present invention proposes a full-degree-of-freedom permanent magnet active construction method based on a deep neural network, which reduces the restrictions on the geometric model of the magnet as much as possible, and uses a deep reinforcement learning algorithm to actively explore the arrangement of magnetic blocks, which can A permanent magnet model with high field strength and good uniformity is constructed under the condition of light weight constraints.
4、本发明基于深度神经网络的输入不仅包括磁块排布状态，还包括磁场状态，两个状态统称为磁体状态，能够全方位多角度描述磁体特性，两种信息的交互与融合提升了磁块动作决策网络的性能。4. The input of the present invention based on the deep neural network includes not only the arrangement state of the magnetic blocks, but also the state of the magnetic field. The two states are collectively referred to as the state of the magnet, which can describe the characteristics of the magnet in all directions and from multiple angles. The interaction and fusion of the two kinds of information improves the magnetic field. Performance of Block Action Decision Networks.
5、本发明构建的基于有限元分析的磁场评价环境具有剩磁退磁效应建模能力，能够准确评价磁场，给出真实且稳定的奖励值反馈，提升磁块动作决策网络训练的稳定性与客观性。5. The magnetic field evaluation environment based on finite element analysis constructed by the present invention has the modeling ability of remanence and demagnetization effects, can accurately evaluate the magnetic field, give real and stable reward value feedback, and improve the stability and objectivity of magnetic block action decision-making network training sex.
6、本发明构建的全自由度磁体几何模型具有位置自由、方向自由、层数自由等特点，极大提升了磁体几何模型的理论最优性能，给磁块动作决策网络提供了更大的优化空间。6. The full-degree-of-freedom magnet geometric model constructed by the present invention has the characteristics of position freedom, direction freedom, and layer number freedom, which greatly improves the theoretical optimal performance of the magnet geometric model and provides greater optimization for the magnetic block action decision network space.
附图说明Description of drawings
图1是本发明的一种全自由度永磁体主动构建方法流程图；Fig. 1 is a flow chart of the active construction method of a full-degree-of-freedom permanent magnet of the present invention;
图2是本发明的全自由度磁体几何模型示意图。Fig. 2 is a schematic diagram of a geometric model of a full-degree-of-freedom magnet of the present invention.
具体实施方式detailed description
以下结合附图对本发明实施例作进一步详述：Embodiments of the present invention are described in further detail below in conjunction with the accompanying drawings:
一种全自由度永磁体主动构建方法，如图1和图2所示，包括以下步骤：An active construction method for a full-degree-of-freedom permanent magnet, as shown in Figure 1 and Figure 2, includes the following steps:
步骤1、选取永磁材料并制作磁块；
所述步骤1的具体方法为：The concrete method of described
根据磁能积、剩磁温度系数与可加工性选取永磁材料，并根据需求制作不同几何尺寸的磁块；Select permanent magnet materials according to magnetic energy product, remanence temperature coefficient and machinability, and make magnets of different geometric sizes according to requirements;
在本实施例中，根据磁能积、剩磁温度系数与可加工性选取永磁材料，并根据需求制作不同几何尺寸的磁块，如1英寸边长的立方体；In this embodiment, permanent magnet materials are selected according to magnetic energy product, temperature coefficient of remanence and machinability, and magnetic blocks of different geometric sizes are made according to requirements, such as a cube with a side length of 1 inch;
步骤2、构建全自由度磁体几何模型；Step 2. Construct a full-degree-of-freedom magnet geometric model;
所述步骤2的具体方法为：The concrete method of described step 2 is:
构建全自由度磁体几何模型，在该磁体几何模型中，磁块的方向(朝向)自由、磁块在模型中的候选位置自由、磁体参数自由。其中磁体参数包括行、列和层数，“行”代表磁体轴向的“磁块环”、“列”代表与磁体轴向平行的“磁块列”、相同孔径的所有相邻列共同组成“层”。A full-degree-of-freedom magnet geometric model is constructed. In the magnet geometric model, the direction (orientation) of the magnet block is free, the candidate position of the magnet block in the model is free, and the magnet parameters are free. Among them, the magnet parameters include row, column and layer number, "row" represents the "magnetic block ring" in the axial direction of the magnet, "column" represents the "magnetic block column" parallel to the magnet axial direction, and all adjacent columns with the same aperture are jointly composed "Floor".
在本实施例中，构建全自由度磁体几何模型，在该磁体几何模型中，磁块的方向(朝向)自由，磁块在模型中的候选位置，磁体参数自由。其中磁体参数包括行、列和层数，具体几何模型参数为：第一层L1包括16行42列、第2层L2包括14行44列、第三层L3包括12行48列、第四层L4包括8行52列。In this embodiment, a full-degree-of-freedom magnet geometric model is constructed. In the magnet geometric model, the direction (orientation) of the magnet block is free, the candidate position of the magnet block in the model, and the magnet parameters are free. The magnet parameters include rows, columns and layers, and the specific geometric model parameters are: the first layer L1 includes 16 rows and 42 columns, the second layer L2 includes 14 rows and 44 columns, the third layer L3 includes 12 rows and 48 columns, The fourth layer L4 includes 8 rows and 52 columns.
后续的磁块排布优化过程均在此磁体几何模型上进行，用以填充该磁体几何模型的磁块均来自步骤1。Subsequent magnet block layout optimization processes are all carried out on this magnet geometric model, and the magnetic blocks used to fill this magnet geometric model are all from
步骤3、基于步骤2构建的全自由度磁体几何模型，确定由少量磁块组成的磁体初始状态，设定初始磁块为原始Halbach角度；Step 3. Based on the full-degree-of-freedom magnet geometric model constructed in step 2, determine the initial state of the magnet composed of a small number of magnetic blocks, and set the initial magnetic block to the original Halbach angle;
所述步骤3的具体方法为：The concrete method of described step 3 is:
在上述磁体几何模型中，首先确定由少量磁块组成的磁体初始状态，磁体的初始状态包括在磁体几何模型中若干层的若干中间行相对均匀分布的若干磁块，设定初始磁块为原始Halbach角度。In the above geometric model of the magnet, the initial state of the magnet composed of a small number of magnetic blocks is firstly determined. The initial state of the magnet includes a number of relatively evenly distributed magnetic blocks in several middle rows of several layers in the geometric model of the magnet. The initial magnetic block is set as the original Halbach angle.
在本实施例中，在上述磁体几何模型中，首先确定由少量磁块组成的磁体初始状态，磁体的初始状态包括在磁体几何模型中若干层的若干中间行相对均匀分布的若干磁块，如8块，设定初始磁块为原始Halbach角度。In this embodiment, in the above-mentioned magnet geometric model, the initial state of the magnet composed of a small number of magnetic blocks is firstly determined, and the initial state of the magnet includes several magnetic blocks that are relatively evenly distributed in several middle rows of several layers in the magnet geometric model, such as 8 blocks, set the initial magnetic block to the original Halbach angle.
步骤4、在强化学习的框架下，构建磁场评价环境；Step 4. Under the framework of reinforcement learning, construct a magnetic field evaluation environment;
所述步骤4的具体方法为：The concrete method of described step 4 is:
在强化学习的框架下，构建基于有限元分析的包含有剩磁退磁效应建模的磁场评价环境，该磁场评价环境用于实现磁体状态的转移以及磁块动作决策前后成像区域磁场均值与均匀度改善程度的评估；Under the framework of reinforcement learning, a magnetic field evaluation environment including remanence demagnetization effect modeling based on finite element analysis is constructed. This magnetic field evaluation environment is used to realize the transfer of the magnet state and the average and uniformity of the magnetic field in the imaging area before and after the decision-making of the magnetic block. assessment of improvement;
所述均匀度改善程度称为奖励R，可用的评价指标包括平均磁场强度Bmean和磁场均匀度Bhomogeneous等。The improvement degree of uniformity is called reward R, and available evaluation indexes include average magnetic field strength B mean and magnetic field uniformity B homogeneous , etc.
在本实施例中，在强化学习的框架下，构建基于有限元分析的包含有剩磁退磁效应建模的磁场评价环境，磁场评价环境能够实现磁体状态的转移以及磁块动作决策前后成像区域磁场均值与均匀度改善程度的度量。上述改善程度称为奖励R，可用的评价指标包括平均磁场强度Bmean以及磁场均匀度Bhomogeneous等，本实施例中R＝w1×Bmean+w2×Bhomogeneous，w1＝1000，w2＝1。In this embodiment, under the framework of reinforcement learning, a magnetic field evaluation environment based on finite element analysis including modeling of remanence and demagnetization effects is constructed. The magnetic field evaluation environment can realize the transfer of magnet state and the magnetic field in the imaging area before and after the action decision of the magnetic block. A measure of mean and uniformity improvement. The above-mentioned degree of improvement is called reward R, and available evaluation indicators include average magnetic field strength B mean and magnetic field uniformity B homogeneous , etc. In this embodiment, R=w 1 ×B mean +w 2 ×B homogeneous , w 1 =1000,w 2 =1.
步骤5、构建基于深度神经网络的磁块动作决策网络，对磁体状态中的磁场状态及其对应磁块排布状态进行不同方式的特征提取与融合，最终实现磁体状态到Q值的精确映射；Step 5. Construct a decision-making network for magnetic block action based on a deep neural network, perform feature extraction and fusion in different ways on the magnetic field state in the magnet state and the corresponding magnetic block arrangement state, and finally realize the accurate mapping from the magnet state to the Q value;
所述步骤5的具体步骤包括：The concrete steps of described step 5 include:
(1)对于磁场状态，构建深度神经网络对其进行特征提取，得到磁场特征；(1) For the state of the magnetic field, construct a deep neural network for feature extraction to obtain magnetic field features;
(2)对于磁场状态对应的磁块排布状态，用磁块状态向量编码磁块状态，一个磁块状态子向量长度为n+1，表示磁块状态子向量中用1个标量来编码磁块位置是否为空，用n个标量来编码磁块的角度；(2) For the magnetic block arrangement state corresponding to the magnetic field state, the magnetic block state is encoded by the magnetic block state vector, and the length of a magnetic block state subvector is n+1, which means that one scalar is used to encode the magnetic block state subvector. Whether the block position is empty, use n scalars to encode the angle of the magnetic block;
(3)将磁块状态向量输入深度神经网络得到磁块排布特征；(3) Input the magnetic block state vector into the deep neural network to obtain the magnetic block arrangement feature;
(4)将磁场特征与磁块排布特征输入到深度神经网络的特征融合模块进行特征的交互与融合；(4) Input the magnetic field feature and the magnetic block arrangement feature into the feature fusion module of the deep neural network to perform feature interaction and fusion;
(5)在深度神经网络最后的分类层得到磁体状态到Q值的映射，其中Q值在深度强化学习中代表奖励值的累积折扣期望；(5) In the final classification layer of the deep neural network, the mapping of the magnet state to the Q value is obtained, where the Q value represents the cumulative discounted expectation of the reward value in deep reinforcement learning;
在本实施例中，构建基于Transformer的磁块动作决策网络，对磁体状态中的磁场状态及其对应磁块排布状态进行不同方式的特征提取。对于磁场状态，首先将其分解为磁场块，然后将分解的每个磁场块都映射为长度为d＝512的一维向量，称为磁场块向量，磁场块向量具有c1＝1000个；对于磁场状态对应的磁块排布状态，将相邻位置的磁块进行三维分解，形成多个磁块组，用磁块状态向量编码磁块状态，一个磁块状态子向量长度为n+1，其中n＝255，表示磁块状态子向量中用1个标量来编码磁块位置是否为空，用n个标量来编码磁块的角度。将每个磁块组中的磁块状态向量映射为一维向量，称为磁块组向量，磁块组向量具有c2＝8个；此外还设定了一个类别向量。因此共有c1+c2+1＝1009个维度为512的输入向量。为了保持数据状态的空间信息，构建1009个位置向量然后通过相加的方式嵌入到输入向量。最终形成一个512×1009的矩阵。该矩阵被输入6个Transformer编码层中进行不同类型的特征交互以预测相位编码。一个Transformer编码层依顺序包括一个归一化层、一个多头注意力层、一个归一化层和一个多层感知机。最后一个Transformer编码层的输出矩阵维度仍然为512×1009，将输出矩阵中类别向量对应的特征输入一个具有归一化层和全连接层的映射头实现磁体状态到Q值的映射，其中Q值在深度强化学习中代表奖励值的累积折扣期望。Q值的训练过程结合了时间差分和当前奖励值(以t时刻为例，Qt＝Rt+Qt+1)，首先将当前奖励值以时间差分的方式映射为Qtarget；然后根据Qtarget和决策网络预测的Q值计算L1Loss并反馈到决策网络的各层，即可实现磁块动作决策网络的参数优化，使其具备依据当前磁体状态做出高性能磁块动作决策的能力；In this embodiment, a Transformer-based magnetic block action decision-making network is constructed, and feature extraction in different ways is performed on the magnetic field state in the magnet state and the corresponding magnetic block arrangement state. For the magnetic field state, it is firstly decomposed into magnetic field blocks, and then each decomposed magnetic field block is mapped to a one-dimensional vector with a length of d=512, which is called a magnetic field block vector, and the magnetic field block vector has c 1 =1000; for The magnetic block arrangement state corresponding to the magnetic field state, three-dimensionally decompose the adjacent magnetic blocks to form multiple magnetic block groups, and use the magnetic block state vector to encode the magnetic block state. The length of a magnetic block state sub-vector is n+1, Among them, n=255, indicating that one scalar is used to encode whether the position of the magnetic block is empty in the state subvector of the magnetic block, and n scalars are used to encode the angle of the magnetic block. The magnetic block state vector in each magnetic block group is mapped to a one-dimensional vector, which is called a magnetic block group vector, and the magnetic block group vector has c 2 =8; in addition, a category vector is also set. Therefore, there are c 1 +c 2 +1=1009 input vectors with dimensions 512 in total. In order to maintain the spatial information of the data state, 1009 position vectors are constructed and then embedded into the input vector by addition. Finally, a 512×1009 matrix is formed. This matrix is input into 6 Transformer encoding layers for different types of feature interactions to predict phase encoding. A Transformer encoding layer consists of a normalization layer, a multi-head attention layer, a normalization layer, and a multi-layer perceptron in sequence. The output matrix dimension of the last Transformer encoding layer is still 512×1009, and the features corresponding to the category vector in the output matrix are input into a mapping head with a normalization layer and a fully connected layer to realize the mapping from the magnet state to the Q value, where the Q value Representing cumulative discounted expectations of reward values in deep reinforcement learning. The training process of Q value combines the time difference and the current reward value (take time t as an example, Q t = R t + Q t+1 ), firstly, the current reward value is mapped to Q target in the way of time difference; then according to Q The target and the Q value predicted by the decision-making network calculate L1Loss and feed it back to each layer of the decision-making network, so as to realize the parameter optimization of the magnetic block action decision-making network, so that it has the ability to make high-performance magnetic block action decisions based on the current magnet state;
步骤6、基于步骤4构建的磁体评价环境和步骤5构建的基于深度神经网络的磁块动作决策网络，依据步骤3预定的初始状态与设定的重量约束条件进行磁块动作决策网络与磁场评价环境的交互，将交互产生的经验保存到回放单元；Step 6. Based on the magnet evaluation environment built in step 4 and the magnet block action decision network based on the deep neural network built in step 5, the magnet block action decision network and magnetic field evaluation are performed according to the initial state predetermined in step 3 and the set weight constraints The interaction of the environment saves the experience generated by the interaction to the playback unit;
所述步骤6的具体步骤包括：The concrete steps of described step 6 include:
(1)在步骤3预定的磁体初始状态下，以磁体重量W为终止条件进行步骤5所述磁块动作决策网络与步骤4所述磁场评价环境的交互，在达到重量上限后返回步骤3所述磁体初始状态并重复交互过程；(1) In the initial state of the magnet predetermined in step 3, the magnet weight W is used as the termination condition to carry out the interaction between the magnet block action decision network described in step 5 and the magnetic field evaluation environment described in step 4, and return to step 3 after reaching the upper limit of weight Describe the initial state of the magnet and repeat the interaction process;
(2)将交互过程中产生的磁块动作决策网络与磁场评价环境的互动经验保存到回放单元，每一条经验包括当前磁体数据状态、磁块动作、奖励以及下一个磁体数据状态；(2) Save the interaction experience between the magnetic block action decision network and the magnetic field evaluation environment generated during the interaction process to the playback unit, each experience includes the current magnet data state, magnetic block action, reward and next magnet data state;
在本实施例中，在上述磁体初始状态下，以磁体重量W＝120kg为终止条件进行磁块动作决策网络与磁场评价环境的交互，在达到重量上限后返回磁体初始状态并重复交互过程。将每一步磁块动作决策网络与磁场评价环境的互动经验保存到回放单元，每一条经验包括当前磁体数据状态、磁块动作、奖励以及下一个磁体数据状态；In this embodiment, in the initial state of the magnet, the interaction between the magnet block action decision network and the magnetic field evaluation environment is performed with the magnet weight W=120kg as the termination condition, and the initial state of the magnet is returned after reaching the upper limit of weight and the interaction process is repeated. Save the interaction experience between the magnetic block action decision-making network and the magnetic field evaluation environment at each step to the playback unit, each experience includes the current magnet data state, magnetic block action, reward and the next magnet data state;
步骤7、将步骤6所得经验输入到步骤5所述磁块动作决策网络中，对磁块动作决策网络进行优化，依据收敛的永磁体模型制造永磁体；Step 7, input the experience obtained in step 6 into the magnetic block action decision network described in step 5, optimize the magnetic block action decision network, and manufacture permanent magnets according to the converged permanent magnet model;
所述步骤7的具体方法为：The concrete method of described step 7 is:
从步骤6所述回放单元中取出经验对步骤5所述磁块动作决策网络进行优化，当磁块动作决策网络收敛时即视为构建出了最优性能的永磁体模型，根据主动构建的全自由度永磁体模型即可制造主磁体。Take experience from the playback unit described in step 6 to optimize the magnetic block action decision-making network described in step 5. When the magnetic block action decision-making network converges, it is considered to have constructed a permanent magnet model with optimal performance. The main magnet can be manufactured from the permanent magnet model with degrees of freedom.
需要强调的是，本发明所述实施例是说明性的，而不是限定性的，因此本发明包括并不限于具体实施方式中所述实施例，凡是由本领域技术人员根据本发明的技术方案得出的其他实施方式，同样属于本发明保护的范围。It should be emphasized that the embodiments of the present invention are illustrative rather than restrictive, so the present invention includes and is not limited to the embodiments described in the specific implementation, and those who are obtained by those skilled in the art according to the technical solutions of the present invention Other implementation modes mentioned above also belong to the protection scope of the present invention.
Claims (8)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211031315.3A CN115455505B (en) | 2022-08-26 | 2022-08-26 | An active construction method for full-degree-of-freedom permanent magnets |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211031315.3A CN115455505B (en) | 2022-08-26 | 2022-08-26 | An active construction method for full-degree-of-freedom permanent magnets |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115455505A true CN115455505A (en) | 2022-12-09 |
| CN115455505B CN115455505B (en) | 2023-11-24 |
Family
ID=84301453
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211031315.3A Active CN115455505B (en) | 2022-08-26 | 2022-08-26 | An active construction method for full-degree-of-freedom permanent magnets |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115455505B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116702568A (en) * | 2023-08-04 | 2023-09-05 | 天津天达图治科技有限公司 | A magnetic resonance imaging permanent magnet design method, system, equipment and medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106788028A (en) * | 2016-12-20 | 2017-05-31 | 江苏大学 | Bearing-free permanent magnet synchronous motor intensified learning controller and its building method |
| CN112532002A (en) * | 2020-11-12 | 2021-03-19 | 华中科技大学 | Double-stator excitation full-freedom-degree bearingless motor and active control method thereof |
| US20210116888A1 (en) * | 2019-10-21 | 2021-04-22 | Semiconductor Components Industries, Llc | Systems and methods for system optimization and/or failure detection |
| CN113343592A (en) * | 2021-07-28 | 2021-09-03 | 辽宁锐翔通用飞机制造有限公司 | DQN intelligent control method for permanent magnet synchronous motor of new energy airplane |
| CN113364386A (en) * | 2021-05-26 | 2021-09-07 | 潍柴动力股份有限公司 | H-infinity current control method and system based on reinforcement learning of permanent magnet synchronous motor |
| WO2022083029A1 (en) * | 2020-10-19 | 2022-04-28 | 深圳大学 | Decision-making method based on deep reinforcement learning |
| US20220208355A1 (en) * | 2020-12-30 | 2022-06-30 | London Health Sciences Centre Research Inc. | Contrast-agent-free medical diagnostic imaging |
| CN114723840A (en) * | 2022-03-09 | 2022-07-08 | 天津大学 | A slice-adaptive-oriented approach to active undersampling in magnetic resonance imaging |
-
2022
- 2022-08-26 CN CN202211031315.3A patent/CN115455505B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106788028A (en) * | 2016-12-20 | 2017-05-31 | 江苏大学 | Bearing-free permanent magnet synchronous motor intensified learning controller and its building method |
| US20210116888A1 (en) * | 2019-10-21 | 2021-04-22 | Semiconductor Components Industries, Llc | Systems and methods for system optimization and/or failure detection |
| WO2022083029A1 (en) * | 2020-10-19 | 2022-04-28 | 深圳大学 | Decision-making method based on deep reinforcement learning |
| CN112532002A (en) * | 2020-11-12 | 2021-03-19 | 华中科技大学 | Double-stator excitation full-freedom-degree bearingless motor and active control method thereof |
| US20220208355A1 (en) * | 2020-12-30 | 2022-06-30 | London Health Sciences Centre Research Inc. | Contrast-agent-free medical diagnostic imaging |
| CN113364386A (en) * | 2021-05-26 | 2021-09-07 | 潍柴动力股份有限公司 | H-infinity current control method and system based on reinforcement learning of permanent magnet synchronous motor |
| CN113343592A (en) * | 2021-07-28 | 2021-09-03 | 辽宁锐翔通用飞机制造有限公司 | DQN intelligent control method for permanent magnet synchronous motor of new energy airplane |
| CN114723840A (en) * | 2022-03-09 | 2022-07-08 | 天津大学 | A slice-adaptive-oriented approach to active undersampling in magnetic resonance imaging |
Non-Patent Citations (2)
| Title |
|---|
| VAN TAI NGUYEN, ET AL.: "Efficient modelling of permanent magnet field distribution for deep learning applications", 《JOURNAL OF MAGNETISM AND MAGNETIC MATERIALS 》 * |
| 王咏涛: "一种新型永磁球形多自由度电机的研究", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116702568A (en) * | 2023-08-04 | 2023-09-05 | 天津天达图治科技有限公司 | A magnetic resonance imaging permanent magnet design method, system, equipment and medium |
| CN116702568B (en) * | 2023-08-04 | 2023-11-10 | 天津天达图治科技有限公司 | A magnetic resonance imaging permanent magnet design method, system, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115455505B (en) | 2023-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113053115B (en) | Traffic prediction method based on multi-scale graph convolution network model | |
| CN112396181A (en) | Automatic pruning method and platform for general compression architecture of convolutional neural network | |
| CN109241313A (en) | A kind of image search method based on the study of high-order depth Hash | |
| CN103473465B (en) | Land resource spatial configuration optimal method based on multiple target artificial immune system | |
| CN112766687B (en) | Aircraft final assembly material distribution priority distribution method based on depth CNN | |
| CN113297174A (en) | Land use change simulation method based on deep learning | |
| CN109902714A (en) | A Multimodal Medical Image Retrieval Method Based on Multi-Graph Regularized Deep Hashing | |
| CN114693064B (en) | Building group scheme generation performance evaluation method | |
| CN115455505B (en) | An active construction method for full-degree-of-freedom permanent magnets | |
| CN115907122A (en) | Regional electric vehicle charging load prediction method | |
| CN113935556B (en) | Temperature sensor optimal arrangement method based on DNA genetic algorithm | |
| CN118196600A (en) | Neural architecture searching method and system based on differential evolution algorithm | |
| CN114647758A (en) | Video abstract generation network based on Transformer and deep reinforcement learning | |
| CN116522712A (en) | Geothermal heat storage temperature field simulation method of neural network | |
| CN118658060A (en) | A hyperspectral image classification method based on agent-assisted multi-objective evolutionary neural architecture search | |
| CN116977723A (en) | Hyperspectral image classification method based on space-spectrum hybrid self-attention mechanism | |
| WO2024011979A1 (en) | Pre-trained deep neural network based superconducting material prediction method and apparatus | |
| Wang et al. | An automated learning method of semantic segmentation for train autonomous driving environment understanding | |
| CN115908909A (en) | Evolutionary Neural Architecture Search Method and System Based on Bayesian Convolutional Neural Network | |
| CN117975174B (en) | Three-dimensional digital core reconstruction method based on improvement VQGAN | |
| CN115762675B (en) | A method and system for predicting mechanical properties of truss-type lattice structures | |
| CN109033521B (en) | Newly-built railway slope-limiting optimization decision method | |
| CN114254803A (en) | Reloading loading scheme search optimization method | |
| CN115130386A (en) | A fast one-to-many design method for digital metasurfaces based on nested multilayer perceptrons | |
| Yu et al. | Locality sensitive hashing-driven multifactorial evolutionary algorithms for multitask optimization |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |