CN111651936A - A FOA-GRNN-based modeling and design method for dual-notch ultra-wideband antennas - Google Patents

Abstract

The invention provides a FOA-GRNN-based dual-notch characteristic ultra-wideband antenna modeling design method, and relates to the technical field of radio frequency microwave device neural network modeling. And predicting the miniaturized ultra-wideband antenna with the double-notch characteristic by using an FOA-GRNN model, training the model by using data of return loss | S11| and frequency f of the antenna, and calculating the relative error of predicted output data of the model to be compared with the GRNN network. The method has better prediction capability, high convergence speed and higher precision.

Description

A FOA-GRNN-based modeling and design method for dual-notch ultra-wideband antennas

technical field

The invention relates to the technical field of neural network modeling of radio frequency microwave devices, in particular to a modeling and design method for an ultra-wideband antenna with double-notch characteristics based on FOA-GRNN.

Background technique

With the rapid development of today's mobile communication, wireless local area network, radio frequency identification, etc., wireless communication systems show great potential, and the design of radio frequency microwave devices directly affects the performance of the entire communication system.

Antenna plays an indispensable role in wireless communication technology. It is the core component that transmits and receives wireless electromagnetic waves and completes electromagnetic conversion. Among them, the ultra-wideband notch antenna has the advantages of small size, wide coverage bandwidth, high radiation efficiency, and can avoid the interference of other communication system frequency bands such as WIMAX (3.3-3.6GHz) and WLAN (5.2-5.8GHz). in ultra-wideband wireless communication systems. In the design of the ultra-wideband notch antenna, the size and position of the resonator can be changed, or the length and width of the slot on the radiation patch can be changed to realize the notch characteristic of the antenna. The commonly used antenna design and analysis method adopts the aided design of HFSS electromagnetic simulation software. When designing a device, once a certain physical size of the antenna needs a slight change, the design needs to be re-optimized, and an electromagnetic simulation process needs to be re-completed, which reduces the design speed and increases the design difficulty.

SUMMARY OF THE INVENTION

In view of the problems existing in the prior art, the present invention provides a FOA-GRNN-based modeling and design method for an ultra-wideband antenna with dual-notch characteristics. The radio frequency characteristics of an antenna with double-notch characteristics are nonlinear correlations. The data of return loss |S11| and frequency f of the notch ultra-wideband antenna are used as sample input, the smooth factor of the generalized regression neural network is optimized by the fruit fly algorithm, and the root mean square error function is used as the fitness function to find the optimal parameters for modeling design.

The technical scheme of the present invention is:

A method for modeling and designing an ultra-wideband antenna with dual-notch characteristics based on FOA-GRNN, comprising the following steps:

Step 1: Select the sample object; select an ultra-wideband antenna with double notch characteristics, and use the relationship data of its return loss |S11| and frequency f as the input sample of the model;

The model is to use electromagnetic simulation software to carry out the simulation modeling of the radio frequency characteristics of the ultra-wideband antenna with double-notch characteristics;

Step 2: Basic parameter settings; set the sizepop of the fruit fly population, the maximum number of iteration steps maxgen, and the coordinate values of the initial position of the fruit fly (X_axis, Y_axis);

Step 3: Extract the training set data; divide the input sample data into two groups of data, namely the training set data and the test set data, perform cross-training of the generalized regression neural network, and use the root mean square error function of the predicted samples as the fitness function;

The input sample data is divided into two groups of data, the two groups of data are alternately used as training set data and test set data, and the generalized regression neural network, that is, the GRNN network structure, is cross-trained to avoid overtraining of the algorithm;

The root mean square error RMSE of the predicted sample is used as the fitness function, as follows:

为广义回归神经网络预测输出值，y_l为实际目标输出值，where M is the number of test samples,

is the predicted output value of the generalized regression neural network, y _l is the actual target output value,

l = 1, 2, ..., M;

Step 4: Optimize the smooth factor of training generalized regression neural network with Drosophila algorithm;

The step 4 specifically includes:

Step 4.1: Randomly set the initial fly direction and distance of the fruit fly;

其中(X(i)，Y(i))为果蝇的位置坐标值，其中i表示果蝇种群数量，i＝1,2,…sizepop，并计算味道浓度判定值S(i)，即为距离之倒数S(i)＝1/D(i)；Step 4.2: Calculate the distance between the fly and the origin

Where (X(i), Y(i)) is the position coordinate value of the fruit fly, where i represents the population of the fruit fly, i=1,2,...sizepop, and calculate the taste concentration judgment value S(i), which is The inverse of the distance S(i)=1/D(i);

Step 4.3: Use the root mean square error of the GRNN network to predict the sample as the fitness fitness function, and substitute the taste concentration judgment value S(i) obtained in step 4.2 into the fruit fly taste concentration value;

Step 4.4: In the taste concentration value obtained in step 4.3, find the minimum value of the taste concentration in the fruit fly population, which is the best taste concentration value bestSmell, and retain the individual position at this time;

Step 4.5: Drosophila iterative optimization starts, compare the taste concentration value of each generation; if the current taste concentration value is smaller than the previous generation, update the taste concentration value to the optimal value;

Step 4.6: Whether the maximum number of iteration steps is met is the most important condition for judging whether the iteration termination is reached. If so, terminate the iteration, record the optimal fruit fly, obtain the optimal smooth factor, and perform FOA-GRNN network modeling; otherwise, return to the execution step 4.1;

Step 5: Finally, the FOA-GRNN network model obtained in the above steps is used in the simulation design of the ultra-wideband antenna with double notch characteristics, and the FOA-GRNN network model is tested with the data that did not participate in the model training, and the obtained output The result is the return loss characteristic data corresponding to the antenna.

The beneficial effects of the present invention are:

The invention provides a FOA-GRNN-based dual-notch characteristic ultra-wideband antenna modeling design method. The fruit fly algorithm is used to optimize the smooth factor of the key GRNN network, and the established FOA-GRNN model has a good prediction effect. The prediction accuracy and fitting ability are obviously superior to the GRNN neural network structure. Finding a suitable modeling method and its fast algorithm can shorten the design cycle, simplify its design process, improve the speed and accuracy of CAD software simulation, and have excellent nonlinear mapping. capability, which provides a convenient method for antenna simulation design.

Description of drawings

Fig. 1 is the flow chart of the dual-notch characteristic ultra-wideband antenna model based on FOA-GRNN of the present invention;

Fig. 2 is the return loss curve diagram of the ultra-wideband antenna of the double-notch characteristic of the embodiment of the present invention;

Fig. 3 is the comparison graph of the predicted output and the actual output of the double-notch characteristic ultra-wideband antenna model of the GRNN according to the embodiment of the present invention;

Fig. 4 is the comparison graph of the predicted output and the actual output of the double-notch characteristic ultra-wideband antenna model of FOA-GRNN according to the embodiment of the present invention;

5 is a comparison diagram of two kinds of structure prediction outputs according to an embodiment of the present invention;

Fig. 6 is the prediction relative error curve diagram of the dual-notch characteristic ultra-wideband antenna model of the GRNN according to the embodiment of the present invention;

Fig. 7 is the prediction relative error curve diagram of the dual-notch characteristic ultra-wideband antenna model of FOA-GRNN according to an embodiment of the present invention;

8 is a flight trajectory diagram of a fruit fly according to an embodiment of the present invention;

FIG. 9 is a diagram of an optimization process of the fruit fly algorithm according to an embodiment of the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

A FOA-GRNN-based dual-notch characteristic ultra-wideband antenna modeling and design method, on the MATLAB platform, establishes a FOA-GRNN network model, and its overall flow chart is shown in Figure 1, including the following steps:

Step 1: Select the sample object; select an ultra-wideband antenna with double notch characteristics, and use the relationship data of its return loss |S11| and frequency f as the input sample of the model;

The model is to use the electromagnetic simulation software HFSS15.0 simulation software to simulate the radio frequency characteristics of the ultra-wideband antenna with double notch characteristics; when the return loss is less than -10dB, the antenna bandwidth is 2.8-12.3GHz, and the frequency is set. The step size is 0.1 GHz, and 140 groups of data of antenna return loss |S11| and frequency f are extracted.

In this embodiment, the electromagnetic simulation software HFSS15.0 is used to model and simulate the radio frequency characteristics of the ultra-wideband antenna with miniaturized double-notch characteristics. As shown in Figure 2, the bandwidth of the antenna return loss less than -10dB is 2.8-12.3GHz, and it produces better notch characteristics in the two frequency bands of 3.23-3.7GHz and 8.01-8.66GHz. Extract the data of its radio frequency characteristic curve, the frequency step is 0.1GHz, and import it into GRNN neural network and FOA-GRNN neural network model as input sample data respectively.

Step 2: Basic parameter settings; set the sizepop of the fruit fly population, the maximum number of iteration steps maxgen, and the coordinate values of the initial position of the fruit fly (X_axis, Y_axis);

Step 3: Extract the training set data; divide the input sample data into two groups of data, namely the training set data and the test set data, perform cross-training of the generalized regression neural network, and use the root mean square error function of the predicted samples as the fitness function;

The input sample data is divided into two groups of data, each group of data is 56 groups of data, and the two groups of data are alternately used as training set data and test set data, and cross-training generalized regression neural network, that is, GRNN network structure, to avoid algorithm overtraining .

The root mean square error RMSE of the predicted sample is used as the fitness function, as follows:

其中

in

为广义回归神经网络预测输出值，y_l为实际目标输出值，where M is the number of test samples,

is the predicted output value of the generalized regression neural network, y _l is the actual target output value,

l = 1, 2, ..., M;

Step 4: Optimize the smooth factor of training generalized regression neural network with Drosophila algorithm;

The step 4 specifically includes:

Step 4.1: Randomly set the initial fly direction and distance of the fruit fly;

In this embodiment, the random flight direction and distance are (-10, 10).

X'(i)=X_axis+20*rand()-10

Y'(i)=Y_axis+20*rand()-10

Where X'(i), Y'(i) represent the transformation direction of the fruit fly's position coordinates; rand() is any random number between [0,1].

其中(X(i)，Y(i))为果蝇的位置坐标值，其中i表示果蝇种群数量，i＝1,2,…sizepop，并计算味道浓度判定值S(i)，即为距离之倒数S(i)＝1/D(i)；Step 4.2: Calculate the distance between the fly and the origin

Where (X(i), Y(i)) is the position coordinate value of the fruit fly, where i represents the population of the fruit fly, i=1,2,...sizepop, and calculate the taste concentration judgment value S(i), which is The inverse of the distance S(i)=1/D(i);

Step 4.3: Use the root mean square error of the GRNN network to predict the sample as the fitness fitness function, and substitute the taste concentration judgment value S(i) obtained in step 4.2 into the fruit fly taste concentration value;

Smell(i)=Function(S(i))

Where Smell(i) is the taste concentration corresponding to the i-th fruit fly, Function(S(i)) fitness function;

Step 4.4: In the taste concentration value obtained in step 4.3, find the minimum value of the taste concentration in the fruit fly population, which is the best taste concentration value bestSmell, and retain the individual position at this time;

[bestSmell bestindex]=min(Smell)

Among them, bestindex is the position corresponding to the best taste concentration value bestSmell, and min(Smell) represents the minimum root mean square error, that is, the minimum value of taste concentration;

Step 4.5: Drosophila iterative optimization starts, compare the taste concentration value of each generation; if it is better than the previous generation, record the current best taste concentration value bestSmell and position coordinates;

Step 4.6: Whether the maximum number of iteration steps is met is the most important condition for judging whether the iteration termination is reached. If so, terminate the iteration, record the optimal fruit fly, and substitute the optimal smooth factor obtained by optimization into the trained GRNN network, and input 60 A set of test samples is used to predict the FOA-GRNN model. Analyze the test results and conduct a comparative study with the GRNN network prediction model to evaluate the performance of the model; otherwise, return to step 4.1;

Step 5: Finally, the FOA-GRNN network model obtained in the above steps is used in the simulation design of the ultra-wideband antenna with double notch characteristics, and the FOA-GRNN network model is tested with the data that did not participate in the model training, and the obtained output The result is the return loss characteristic data corresponding to the antenna.

The GRNN neural network and the FOA-GRNN neural network use the same training set and test set, and simultaneously conduct neural network simulation on the MATLAB platform to obtain the prediction results. Figure 3 shows the comparison curve between the predicted output and the actual output of the GRNN network. Figure 4 shows the comparison curve between the predicted output and the actual output of the FOA-GRNN network. Figure 5 is a comparison diagram of the prediction outputs of the two structures. It can be seen by comparison that FOA-GRNN has a better effect on nonlinear approximation.

Part of the predicted values of the two neural networks are shown in Table 1. It can be seen that the prediction error of GRNN is greater than 1%, while the prediction error of FOA-GRNN is basically below 1%, which is closer to the actual data.

Table 1 Part of the comparison table of measured and predicted values

Relative error = (predicted value - measured value) / measured value * 100%

See Fig. 6 for the prediction relative error curve of GRNN. See Fig. 7 for the prediction relative error curve of FOA-GRNN. It can be concluded that the prediction accuracy of the FOA-GRNN model is higher.

See Figure 8 for the flight trajectory of the fruit fly, and Figure 9 for the optimization process of the fruit fly algorithm. It can be seen from the figure that the optimization process begins to converge in the 8th generation of the iteration and ends at the 42nd generation. The root mean square error value at this time is 0.6667. The root mean square error of the GRNN prediction data is 2.0492. It can be seen that the FOA-GRNN neural network has a strong advantage in nonlinear prediction.

Obviously, the above-mentioned embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. The above embodiments are only used to explain the present invention, and do not constitute a limitation on the protection scope of the present invention. Based on the above-mentioned embodiments, all other embodiments obtained by those skilled in the art without creative work, that is, all modifications, equivalent replacements and improvements made within the spirit and principle of the present application, are fall within the scope of protection claimed by the present invention.

Claims (2)

1. A dual-notch characteristic ultra-wideband antenna modeling design method based on FOA-GRNN is characterized by comprising the following steps:

step 1: selecting a sample object; selecting an ultra-wideband antenna with a double-notch characteristic, and using the relation data of return loss | S11| and frequency f of the ultra-wideband antenna as an input sample of a model;

the model is a simulation modeling of radio frequency characteristics of the ultra-wideband antenna with the double trapped wave characteristics by utilizing electromagnetic simulation software;

step 2: setting basic parameters; setting a fruit fly population scale sizepop, a maximum iteration step number maxgen and fruit fly initial position coordinate values (X _ axis, Y _ axis);

and step 3: extracting training set data; dividing input sample data into two groups of data, namely training set data and test set data, performing cross training on the generalized regression neural network, and taking a root mean square error function of a prediction sample as a fitness function;

dividing input sample data into two groups of data, alternately taking the two groups of data as training set data and test set data, and cross-training a Generalized Recurrent Neural Network (GRNN) structure to avoid excessive algorithm training;

the root mean square error RMSE of the prediction samples is taken as a fitness function, and the fitness function comprises:

where M is the number of test samples,

predicting output value, y, for a generalized recurrent neural network_lIs the actual target output value, l ═ 1,2, …, M;

and 4, step 4: optimizing and training a smoothing factor of the generalized regression neural network by using a drosophila algorithm;

and 5: and finally, the FOA-GRNN network model obtained in the step is used in the simulation design of the ultra-wideband antenna with the double-notch characteristic, the FOA-GRNN network model is tested by using data which do not participate in model training, and the obtained output result is the return loss characteristic data corresponding to the antenna.

2. The FOA-GRNN-based modeling and design method for the ultra-wideband antenna with the double notch characteristics as claimed in claim 1, wherein: the step 4 specifically includes:

step 4.1: randomly setting the flight direction and distance of the initial fruit flies;

step 4.2: calculating the distance between the fruit fly and the origin

Wherein (x), (i), y (i)) are coordinates of the location of the drosophila, wherein i represents the population number of drosophila, i is 1,2, … sizepop, and the taste concentration determination value s (i) is calculated as the reciprocal of the distance s (i) is 1/d (i);

step 4.3: substituting the taste concentration judgment value S (i) obtained in the step 4.2 by using the root mean square error of the GRNN network prediction sample as a Fitness Fitness function to obtain a fruit fly taste concentration value;

step 4.4: finding out the minimum value of the taste concentration in the fruit fly population from the taste concentration values obtained in the step 4.3, namely the optimal taste concentration value bestSmell, and keeping the position of the individual at the moment;

step 4.5: starting fruit fly iterative optimization, and comparing the taste concentration values of each generation; if the current generation taste concentration value is smaller than the previous generation, updating the taste concentration value to be an optimal value;

step 4.6: judging whether the maximum iteration step number is met or not to be the condition for judging whether the iteration is terminated or not, if so, terminating the iteration, recording the optimal fruit fly to obtain the optimal smooth factor, and carrying out FOA-GRNN network modeling; otherwise, the step 4.1 is executed.

Images