CN112231986B - Numerical control attenuator modeling method - Google Patents
Numerical control attenuator modeling method Download PDFInfo
- Publication number
- CN112231986B CN112231986B CN202011215028.9A CN202011215028A CN112231986B CN 112231986 B CN112231986 B CN 112231986B CN 202011215028 A CN202011215028 A CN 202011215028A CN 112231986 B CN112231986 B CN 112231986B
- Authority
- CN
- China
- Prior art keywords
- neural network
- numerical control
- data
- attenuation
- control attenuator
- 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.)
- Active
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
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Measurement Of Resistance Or Impedance (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a modeling method of a numerical control attenuator, which comprises the steps of obtaining test data of a reference state and an attenuation ground state, obtaining data of other attenuation states, preprocessing sample data, establishing a neural network model and training the neural network model. And then, a neural network model is constructed based on a BP neural network method by utilizing data in each working state, and the complex nonlinear mapping relation of the model can be accurately fitted. The numerical control attenuator modeling provided by the invention can carry out self-adaptive training on input and output data to learn the behavior characteristics of the device without knowing the internal structure and working principle of a chip, and has obvious advantages compared with the traditional modeling method.
Description
Technical Field
The invention relates to the technical field of radio frequency/microwave modeling simulation, in particular to a modeling method of a numerical control attenuator.
Background
The radio frequency devices are various in types, structures are varied, working states and working principles are different, and the radio frequency devices can be divided into single-state devices and multi-state devices according to the working states of the devices. The numerical control attenuator controls the switching devices at all levels by using codes, so that the branches of each attenuation module are controlled to be in a straight-through state or an attenuation state, and different working states are combined, and the numerical control attenuator is a multi-state chip as shown in fig. 2.
The modeling of the radio frequency device is used as an important link of radio frequency/microwave modeling and simulation, and plays a very important role in accurately evaluating the electrical performance of a radio frequency microwave system, the transmission performance of signals and the like.
The traditional radio frequency chip modeling method mainly comprises the following steps: physical model method, equivalent circuit model method and numerical method. However, the physical model and the equivalent circuit model have fast calculation speed, but the internal structure and the working principle of the chip must be relatively known, and when the parasitic effect and the coupling effect are not deeply known, an accurate model is difficult to obtain; the numerical method requires a huge amount of calculation to obtain a more accurate model.
In addition, for a multi-state device, the input signal of the multi-state device also comprises a control signal besides a radio frequency signal, so that a plurality of working states can be combined, and the test work of the device and the model description of the device are more complicated.
Disclosure of Invention
Aiming at the limitations of the traditional modeling method and the characteristics of multiple states of the numerical control attenuator, the invention provides the numerical control attenuator modeling method, when the method obtains modeling sample data, all working states of the numerical control attenuator are not required to be tested, only a reference state (0 dB attenuation state) and a plurality of attenuation ground states of the numerical control attenuator are required to be tested, and data of other working states are calculated through a de-embedding and cascading algorithm, so that the testing workload can be obviously reduced. And then, a neural network model is constructed based on a BP neural network method by utilizing data in each working state, and the complex nonlinear mapping relation of the model can be accurately fitted. The numerical control attenuator modeling provided by the invention can perform self-adaptive training on input and output data to learn the behavior characteristics of the device without knowing the internal structure and working principle of a chip, and has obvious advantages compared with the traditional modeling method.
In order to achieve the purpose, the numerical control attenuator modeling method comprises the following steps:
step one, obtaining reference state and attenuation ground state test data: aiming at the modeling of the numerical control attenuator, taking the S parameter as an output parameter of a neural network model and taking the frequency as an input parameter; carrying out data measurement on a reference state and each attenuation ground state of the numerical control attenuator by using a vector network analyzer, wherein the attenuation ground state is an attenuation state in which each attenuation module in the numerical control attenuator acts independently;
step two, acquiring data of other attenuation states: solving the S parameters of each attenuation module by a de-embedding formula, and then solving the cascade S parameters under different combinations of a reference state and a plurality of attenuation modules by using a cascade formula to obtain S parameter data under different working states;
step three, sample data preprocessing: carrying out reverse folding processing on periodically changed phase data in the S parameter, and carrying out normalization and random sequencing processing on a neural network training sample;
step four, establishing a neural network model: the data of the numerical control attenuator is autonomously learned by using a BP neural network algorithm to establish a model of the numerical control attenuator, the neural network model is respectively established aiming at S parameters of each dimensionality of the numerical control attenuator in each working state, and the number of layers of each neural network, the number of neurons of each hidden layer and an activation function of each neuron are determined;
step five, training a neural network model: and respectively training the neural network model according to the amplitude and the phase of each dimension S parameter of each working state.
Furthermore, in the first step, the amplitude and the phase of the S parameter of each channel are independently used as the output of the neural network for modeling so as to improve the accuracy and the training efficiency of the model, the model in each working state comprises a plurality of neural network models, and each neural network model only calculates and outputs one-dimensional S parameters.
Further, the reverse folding treatment in the third step comprises the following steps: when the phase data of one point is different from the phase data of the previous point by more than half of the phase period, the phase data of the current point is added or subtracted by a plurality of times of the period value, so that the change of the phase data is relatively smooth, and the correctness of the phase data is not influenced.
Furthermore, the normalization processing in the third step is to normalize the original data to the same scale range, so as to avoid that the absolute value difference of the physical quantities of all dimensions is too large to influence the model precision.
Further, a BP neural network model algorithm is adopted in the fifth step, the training process mainly comprises a forward calculation process and a reverse error propagation process, and the training process is carried out by utilizing a third-party neural network algorithm library.
Further, the neural network training process mainly needs to determine three parameters: training algorithms, learning rates, and iteration termination conditions.
The invention has the beneficial effects that:
according to the modeling method of the numerical control attenuator, modeling data of all other working states can be obtained through the de-embedding and cascading algorithm only by testing the reference state and the attenuation ground states, and the task load of device testing work can be effectively reduced. The invention accurately models the obtained modeling data by using the BP neural network algorithm, does not need deep knowledge on the internal structure and the working principle of a plurality of devices, and can quickly convert the modeling data into the neural network model of the numerical control attenuator without excessive priori knowledge. Based on the modeling method provided by the invention, the testing and modeling efficiency of the numerical control attenuator can be effectively improved.
Drawings
FIG. 1 is a flow chart of the numerical control attenuator modeling of the present invention;
FIG. 2 is a schematic diagram of the functional composition of the numerical control attenuator model of the present invention;
FIG. 3 is a schematic diagram of two-port device cascades and an equivalent network according to the present invention;
4-11 comparison curves of the output and test data of the S parameter neural network model of each dimension in the 9dB attenuation state model of the numerically controlled attenuator according to the embodiment of the invention.
Detailed Description
In order to more clearly understand the technical features, objects, and effects of the present invention, specific embodiments of the present invention will now be described. It should be understood that the detailed description and specific examples, while indicating embodiments of the invention, are given by way of illustration only, not by way of limitation, i.e., the embodiments described are intended as a selection of the best mode contemplated for carrying out the invention, not as a full mode. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a modeling method of a numerical control attenuator, which comprises the following steps as shown in figure 1:
step one, obtaining reference state and attenuation ground state test data
Aiming at the modeling of the numerical control attenuator, the S parameter (the amplitude and the phase of the S parameter of each channel: S) 11_amp ,S 11_pha ，…S nn_amp ,S nn_pha ) As an output parameter of the neural network model, the frequency is used as an input parameter. It should be noted that although the neural network model may be multi-output, in order to improve the model accuracy and training efficiency, the amplitude and phase of the S parameter of each channel are individually modeled as the output of the neural network, so that the model in each working state will include a plurality of neural network models, and each neural network model only calculates and outputs one-dimensional S parameters. Because the numerical control attenuator comprises a plurality of working states, each working state needs to be modeled separately.
As can be seen from fig. 2, the numerical control attenuator is composed of a reference state and a plurality of attenuation modules in cascade connection, and the on-off of each attenuation module branch is controlled by a switch to realize the combination of different attenuation states. The attenuation state in which each attenuation module acts alone is referred to as an attenuation ground state, and other attenuation operating states can be formed by combining a reference state and different attenuation modules.
In the invention, data measurement is carried out on the reference state and each attenuation ground state only by using the vector network analyzer, and data of other attenuation states are obtained by de-embedding and cascading treatment, so that the task load of test work can be obviously reduced.
Step two, acquiring data of other attenuation states
The numerical control attenuator can be seen as one or twoThe various working states of the port device can be regarded as a cascade of a plurality of two-port devices, as shown in fig. 3. Wherein the S parameter (S) after the cascade composite ) Can be calculated from S parameters (S) of two devices before cascade connection (1) And S (2) ) Calculated by a cascade formula. If S is known composite And S (1) Then S can be obtained by de-embedding formula (2) 。
According to the analysis, each attenuation ground state of the numerical control attenuator can be regarded as a reference state and cascade of each attenuation module, and other attenuation states can be regarded as the reference state and cascade of a plurality of attenuation modules. Therefore, the S parameter of each attenuation module is firstly solved through a de-embedding formula, and then the cascade S parameter under different combinations of the reference state and the plurality of attenuation modules is solved through a cascade formula, so that S parameter data under different working states are obtained.
Step three, preprocessing sample data
Preprocessing sample data comprises the steps of performing anti-folding processing on periodically changed phase data, and performing normalization and random sequencing processing on neural network training samples, and the two steps of preprocessing operations are briefly described below.
1) Reverse folding operation
The S parameter includes periodically varying phase data (S) 11_pha 、S 21_pha 、S 12_pha 、S 22_pha ) For periodically changing phase data, in a data point between two periods, drastic mutation of the data can occur, which is represented as periodic folding of a data curve, and the precision of neural network modeling can be influenced. In order to eliminate this effect, the periodic phase data needs to be processed by reverse folding, and the specific operation method is as follows: when the phase data of one point is different from the phase data of the previous point by more than half of the phase period, the phase data of the current point is added or subtracted with a plurality of times of period values, so that the change of the phase data is relatively smooth, and the correctness of the phase data is not influenced.
2) Normalized, random sort operation
The data range is often very different because the input and output physical quantity units and dimensions are different. The normalization processing is to normalize the original data to the same scale range, so as to avoid influence on model accuracy caused by too large difference of absolute values of physical quantities of all dimensions.
The random ordering is to improve generalization ability of the neural network model and improve prediction accuracy of the model at input points outside the training samples, and the random ordering can be generally realized by program automation.
Step four, establishing a neural network model
The method utilizes the BP neural network algorithm to carry out autonomous learning on the data of the numerical control attenuator to establish an accurate model of the numerical control attenuator, and compared with the traditional modeling method, the method does not need to know the internal structure and the working principle of the numerical control attenuator and does not need excessive prior knowledge. And respectively establishing a neural network model aiming at the S parameter of each dimension of the numerical control attenuator in each working state, and determining the number of layers of each neural network, the number of neurons of each hidden layer and the activation function of each neuron.
Step five, training the neural network model
After the processing of the first step to the third step, the neural network model training can be carried out on each working state of the numerical control attenuator, and in order to improve the model precision and the training efficiency, the neural network model training is respectively carried out aiming at the amplitude and the phase of each dimensionality S parameter of each working state. The BP neural network model algorithm is adopted, the training process mainly comprises a forward calculation process and a reverse error propagation process, and the training process is usually carried out by utilizing a third-party neural network algorithm library. Three parameters are mainly determined in the neural network training process: training algorithm, learning rate, iteration termination condition.
In a preferred embodiment of the present invention, the above process is illustrated by taking a digitally controlled attenuator as an example, and comprises the following steps:
the method comprises the following steps: obtaining reference state and decay ground state test data
The numerical control attenuator in this embodiment is a 4-bit numerical control attenuator, and is composed of an 8dB attenuation module, a 4dB attenuation module, a 2dB attenuation module, and a 1dB attenuation module, which are combined to have 15 attenuation states and 1 reference state (0 dB attenuation state, all attenuation module branches are disconnected). Respectively pair by vector network analyzerTesting a reference state and 4 attenuation basic states (8dB, 4dB,2dB and 1dB) of the numerical control attenuator, selecting 191 groups of samples with the test frequency of 1 GHz-20 GHz and the step of 100MHz, inputting sample data into each frequency point (f), and outputting corresponding S parameters (S) of each channel 11_amp ，S 11_pha ，S 21_amp ，S 21_pha ，S 12_amp ，S 12_pha ，S 22_amp ，S 22_pha )。
Step two: obtaining data of other attenuation states
Through a de-embedding formula and a cascading formula, data in any state can be obtained by using reference state data and attenuation basic state data, and the process of obtaining data in any attenuation state is explained by taking data in a 9dB attenuation state as an example.
Because the 9dB attenuation state can be regarded as being formed by cascading the reference state and the 8dB and 1dB attenuation modules, and the 8dB and 1dB attenuation ground states both contain the reference state, the cascade connection is directly carried out, and the result error can be caused by cascading three reference states, therefore, firstly, the 8dB and 1dB attenuation ground states are respectively embedded with the reference states contained in the attenuation ground states by utilizing a de-embedding formula, and S parameters of the 8dB and 1dB attenuation modules are obtained; then, the reference state, the 8dB attenuation module and the 1dB attenuation module are respectively cascaded by utilizing a cascade formula to obtain cascade S parameters of the 9dB attenuation state.
Step three: preprocessing sample data
Preprocessing the obtained cascade S parameter of the 9dB attenuation state, firstly, periodically changing phase data S 11_pha 、S 21_pha 、S 12_pha 、S 22_pha Performing reverse folding treatment by using a reverse folding formula; then, normalizing the input and output dimension S parameters by using a normalization formula respectively; then, randomly ordering the input and output dimension data, wherein the rules and the sequence of random ordering of the dimensions are consistent so as to ensure that the corresponding relation of input and output is unchanged; and finally, taking 80% of the preprocessed sample data as training samples and 20% of the preprocessed sample data as test samples, and directly applying the preprocessed sample data to training of a neural network model.
Step four: establishing a neural network model
The working states of the numerical control attenuator are respectively modeled, and the modeling process is still illustrated by taking 9dB attenuation state modeling as an example. And respectively establishing a neural network model aiming at each dimensionality S parameter of the 9dB attenuation state, giving an initial value of the neural network parameter, and then correcting when the neural network model is trained. The initial values of the network parameters given here are: the number of network layers is 3, wherein the network layers comprise an input layer, a hidden layer and an output layer; because the network input is frequency and the output is S parameter of a certain dimension, the number of neurons of the input layer and the output layer is 1, and the number of neurons of the hidden layer is set to be 20; the activation function of the hidden layer neuron selects a unipolar Sigmoid function, and the activation function of the output layer neuron selects a linear function.
Step five: training neural network model
Training the neural network model requires determining a training algorithm, an iteration termination condition and a learning rate. For this embodiment, the LM algorithm is selected as the training algorithm, the initial value of the learning rate is selected to be 0.3, the maximum number of iterations is 500, and the expected maximum mean square error is 10 -7 。
Here, the neural network model training of each dimension S parameter in the 9dB attenuation state model is still taken as an example to illustrate that, through multiple iterations and modification of corresponding parameters, a better parameter combination is finally obtained: the number of the network layers is 4, the number of the two hidden layer neurons is 15 and 50 respectively, the activation function of the hidden layer neurons is a unipolar Sigmoid function, the learning rate is 0.1, and other parameters are the same as the settings in the steps. The result comparison curves of the S parameter neural network model output and the test data of each dimension are shown in fig. 4-11, and the mean square error and the iteration number are shown in the following table:
in the embodiment, the modeling of the reference state and other attenuation states of the numerical control attenuator is similar to the modeling process of the 9dB attenuation state, and the models of the numerical control attenuator in various working states can be obtained according to the steps.
The foregoing is illustrative of the preferred embodiments of the present invention, and it is to be understood that the invention is not limited to the precise form disclosed herein and is not to be construed as limited to the exclusion of other embodiments, and that various other combinations, modifications, and environments may be used and modifications may be made within the scope of the concepts described herein, either by the above teachings or the skill or knowledge of the relevant art. And that modifications and variations may be effected by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (4)
1. A modeling method of a numerical control attenuator is characterized by comprising the following steps:
step one, obtaining reference state and attenuation ground state test data: aiming at the modeling of the numerical control attenuator, taking the S parameter as an output parameter of a neural network model and taking the frequency as an input parameter; carrying out data measurement on a reference state and each attenuation ground state of the numerical control attenuator by using a vector network analyzer, wherein the attenuation ground state is an attenuation state in which each attenuation module in the numerical control attenuator acts independently; independently modeling the amplitude and the phase of the S parameter of each channel as the output of a neural network so as to improve the model precision and the training efficiency, wherein the model in each working state comprises a plurality of neural network models, and each neural network model only calculates and outputs one-dimensional S parameters;
step two, acquiring data of other attenuation states: the other attenuation states are formed by combining the reference state of the numerical control attenuator with different attenuation modules, the S parameter of each attenuation module is calculated through a de-embedding formula, and then the cascade S parameter under different combinations of the reference state and the plurality of attenuation modules is calculated by utilizing a cascade formula so as to obtain S parameter data under different working states;
step three, preprocessing sample data: carrying out reverse folding processing on periodically changed phase data in the S parameter, and carrying out normalization and random sequencing processing on a neural network training sample; the reverse folding process comprises the following steps: when the phase data of one point and the phase data of the previous point have a difference exceeding half of the phase period, adding or subtracting a plurality of times of period values to the phase data of the current point, so that the change of the phase data is relatively smooth, and the correctness of the phase data is not influenced;
step four, establishing a neural network model: the data of the numerical control attenuator is autonomously learned by using a BP neural network algorithm to establish a model of the numerical control attenuator, the neural network model is respectively established aiming at S parameters of each dimensionality of the numerical control attenuator in each working state, and the number of layers of each neural network, the number of neurons of each hidden layer and an activation function of each neuron are determined;
step five, training a neural network model: and respectively training the neural network model according to the amplitude and the phase of each dimension S parameter of each working state.
2. The method for modeling a numerical control attenuator according to claim 1, wherein in the third step, the normalization process is to normalize the original data to the same scale range, so as to avoid that the absolute value of the physical quantity of each dimension is too different to affect the model accuracy.
3. The modeling method of the numerical control attenuator according to claim 1 or 2, wherein in the fifth step, a BP neural network model algorithm is adopted, and the training process mainly comprises a forward calculation process and a reverse error propagation process and is performed by utilizing a third-party neural network algorithm library.
4. The method according to claim 3, wherein the neural network training process essentially requires three parameters to be determined: training algorithms, learning rates, and iteration termination conditions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011215028.9A CN112231986B (en) | 2020-11-04 | 2020-11-04 | Numerical control attenuator modeling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011215028.9A CN112231986B (en) | 2020-11-04 | 2020-11-04 | Numerical control attenuator modeling method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112231986A CN112231986A (en) | 2021-01-15 |
| CN112231986B true CN112231986B (en) | 2023-04-07 |
Family
ID=74121869
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011215028.9A Active CN112231986B (en) | 2020-11-04 | 2020-11-04 | Numerical control attenuator modeling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112231986B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446310A (en) * | 2015-08-06 | 2017-02-22 | 新加坡国立大学 | Transistor and system modeling methods based on artificial neural network |
| CN109783905A (en) * | 2018-12-28 | 2019-05-21 | 中国地质大学（武汉） | Microwave Cavity Filter intelligent regulator method based on particle swarm optimization algorithm |
| CN110907785A (en) * | 2018-09-14 | 2020-03-24 | 天津大学青岛海洋技术研究院 | S parameter de-embedding method based on artificial neural network |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787558B (en) * | 2016-04-11 | 2018-07-06 | 江苏科技大学 | Knowledge neural network microstrip filter design method based on ADS |
| CN106411288A (en) * | 2016-08-26 | 2017-02-15 | 吴韵秋 | Multidigit digitally controlled attenuator with low additional phase shift |
| CN108345749B (en) * | 2018-02-11 | 2021-06-15 | 中国电子科技集团公司第二十九研究所 | Modeling and packaging method for radio frequency integrated process tolerance and electrical performance coupling characteristics |
| CN108768550B (en) * | 2018-06-21 | 2021-07-06 | 中国人民解放军国防科技大学 | Wide-band transmitter nonlinear modeling method based on dynamic multi-core bandwidth generalized regression neural network algorithm |
| US10291208B1 (en) * | 2018-07-09 | 2019-05-14 | Psemi Corporation | Method and apparatus for adjusting the slope of insertion loss as a function of frequency of RF digital step attenuators |
| CN109697741B (en) * | 2018-12-28 | 2023-06-16 | 上海联影智能医疗科技有限公司 | PET image reconstruction method, device, equipment and medium |
| CN109921886B (en) * | 2019-01-28 | 2021-08-10 | 东南大学 | Robust low-power-consumption equipment radio frequency fingerprint identification method |
-
2020
- 2020-11-04 CN CN202011215028.9A patent/CN112231986B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446310A (en) * | 2015-08-06 | 2017-02-22 | 新加坡国立大学 | Transistor and system modeling methods based on artificial neural network |
| CN110907785A (en) * | 2018-09-14 | 2020-03-24 | 天津大学青岛海洋技术研究院 | S parameter de-embedding method based on artificial neural network |
| CN109783905A (en) * | 2018-12-28 | 2019-05-21 | 中国地质大学（武汉） | Microwave Cavity Filter intelligent regulator method based on particle swarm optimization algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112231986A (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299430A (en) | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages | |
| CN112231985B (en) | Radio frequency filter modeling method | |
| CN111459022B (en) | Device parameter adjustment method, device control apparatus, and computer-readable storage medium | |
| CN110443724A (en) | A kind of electric system fast state estimation method based on deep learning | |
| CN113919221A (en) | Fan load prediction and analysis method and device based on BP neural network and storage medium | |
| CN109325613B (en) | Method for online predicting static voltage stability margin by using parallel sequential extreme learning machine | |
| CN108460462A (en) | A kind of Interval neural networks learning method based on interval parameter optimization | |
| CN109272144B (en) | BPNN-based prediction method for NDVI (normalized difference of variance) in northern grassland area of China | |
| CN108491931A (en) | A method of non-destructive testing precision is improved based on machine learning | |
| CN106372348A (en) | Vector fitting model order reduction method based on error control in linear system | |
| CN112949944A (en) | Underground water level intelligent prediction method and system based on space-time characteristics | |
| CN113221447A (en) | Soil humidity prediction method for optimizing BP neural network based on improved genetic algorithm | |
| CN106896724B (en) | Tracking system and tracking method for sun tracker | |
| Shang et al. | Research on intelligent pest prediction of based on improved artificial neural network | |
| US20190042681A1 (en) | Method and apparatus for selecting integrated circuit device neural network modeling sample | |
| CN112231986B (en) | Numerical control attenuator modeling method | |
| CN111914488B (en) | Data area hydrologic parameter calibration method based on antagonistic neural network | |
| CN117314236A (en) | BP neural network-based power distribution equipment health state evaluation method and system | |
| CN117077512A (en) | Electromagnetic parametric modeling method for two-port microstrip structure | |
| CN111914465A (en) | Data-free regional hydrological parameter calibration method based on clustering and particle swarm optimization | |
| CN110414146B (en) | Water environment treatment project design parameter optimization method based on deep learning | |
| CN113783186B (en) | Voltage prediction method considering topological structure change of power distribution network | |
| CN109696618A (en) | A kind of adjustment method of radio-frequency devices, device, equipment, storage medium and system | |
| CN112836431A (en) | Penicillin fermentation process fault prediction method based on PSO-LSTM | |
| CN118070688B (en) | Construction method and construction device of power grid load model, storage medium and electronic device |
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 |