main-content

## Weitere Artikel dieser Ausgabe durch Wischen aufrufen

01.12.2019 | Research | Ausgabe 1/2019 Open Access

# The impact mechanism of rural land circulation on promoting rural revitalization based on wireless network development

Zeitschrift:
EURASIP Journal on Wireless Communications and Networking > Ausgabe 1/2019
Autoren:
Ai-min Zheng, Zhong Li
Abbreviations
BP
Backpropagation neural network
Tansig
Tan-sigmoid

## 1 Introduction

The state has a strict scope for the collective use of rural land [1]. Rural land transfer refers to the circulation of collective construction land, which mainly includes farmer’s own housing land, land used by rural self-owned enterprises, and land occupied by rural public places and facilities [2]. Land transfer is a method of land use permitted by national laws. It must be subject to scope and conditional restrictions in order to lease and resell rural collective land. There are many details of the transfer of land in rural areas and the sale of land in cities [3]. The same is true whether the land in the cities or in the countryside is a commodity with circulation value. However, the fact that the land has an immovable position also allows the land circulation to not circulate the land itself. It is actually the change of ownership of the land within a certain period of time. Domestic land transfer and foreign land resale are two different concepts. This is because foreign land is privately owned and domestic land is owned by rural collectives. The impact of the transfer of collectively owned land in China is more complicated than the impact of land transfer abroad. However, in land transfer, the evaluation of land prices is the most critical economic issue affecting land circulation [4]. Under this background, the influencing factors of the valuation of rural collective land are studied. Starting from the actual situation in rural China, it is a subject worthy of advancing to use artificial intelligence algorithm to establish a scientific and objective land evaluation index system. Taking into account the nonlinear characteristics of various policy and social issues, such as the nature of land and market chaos in the process of rural land transfer, neural networks are introduced to study rural land evaluation based on the development of wireless networks. The mechanisms that influence the promotion of rural revitalization are explored [5].

## 2 State of the art

The neural network simulates the human brain’s nervous system and stores and processes information. It has the function of simplifying, summarizing, and simulating human brain information [6]. The neural network learns human behaviors such as learning, memory, inference, and calculation of knowledge. The abstract mathematical model can reflect human understanding of abstract learning and cognitive processes and has played an important role in artificial intelligence research in recent years [7]. The neural network uses a large number of neuron nodes to reflect the structure and function of the human brain, and uses an induction learning method. A large-scale example is used to repeated studies. In the internal process of constant adaptation, the weights of the interconnected neurons are corrected so that the mutual structure and connection weights of the neural networks are distributed stably. The whole process is the learning process of human knowledge acquisition [8]. The neural network can change the method of neural learning and dynamically respond and process external input information through its own neurons, which has the characteristics of storage and application experience knowledge. It is also possible to recognize and remember the characteristics of certain things, and then to distinguish the things based on the stored memory when they encounter the information again later [9]. The neural network uses the system error squared sum as an objective function, which causes the convergence speed to be affected and local minima may occur. The inaccuracy of the prediction of the data model with large gaps in the numerical value relationship has affected the popularization and application in practice. Until the 1980s, after these deficiencies were overcome by scholars, many research results of neural networks were widely used in industrial control, economic research, and engineering construction. The biggest advantages of neural networks are adaptive, nonlinear, and the ability to learn and rectify errors [10].
BP neural network is the main mathematical model for optimization of common neural network models. According to statistical data, it can be seen that 90% of neural network models are now using BP neural networks. It is a multi-layer forward neural network based on error backpropagation algorithm. It can provide a simple nonlinear modeling method for complex systems. It can achieve arbitrary nonlinear mapping near any degree of accuracy [11]. It can autonomously change the internal network through learning. The connection values are actively adapted to the changes of the system, and have better fault tolerance and robustness. The multi-input and output structural model can better use multi-variable system states. To improve the performance level of BP neural network, the flow of the algorithm can be optimized, the convergence period can be used, and the algorithm structure can be updated to improve the evaluation accuracy. Therefore, this paper proposes an algorithm optimization and improvement of the mathematical model, and carries out practical verification in the land evaluation of rural land circulation to examine the improvement of the post-scientific nature of the neural network and the improvement of prediction accuracy [12].

## 3 Methodology

### 3.1 BP neural network

BP neural network is information feed-forward layered neural network. Backward propagation of errors is the most special place for the algorithm. There is also two-way communication in learning. Neural network structure is divided into input layer, output layer, and hidden layer. The hidden layer here may be one layer or n layers. Each layer will forward link through the connection weight between nodes. After a BP neural network enters a signal, the signal is transmitted to the hidden layer. After the excitation function is calculated, the information is passed to the output layer and becomes the output signal. The training algorithm of BP neural network is characterized by two-way propagation. The input layer enters the hidden layer and then the output layer outputs the result. This is the forward state in the two-way propagation. When the hidden layer receives an error between the signal and the output signal, and the error exceeds the desired range, the system corrects the weights and thresholds of the neurons at each layer based on this error, so that each layer becomes more adaptable to promote system performance. This reflects the state of backpropagation. A node in the network corresponds to a single neuron, and there is no intersection between the neurons in different layers. The input of neural nodes at each level only comes from the upper layer that is associated with it. The output of each layer of neurons only has influence on the next layer of neural nodes. In practical applications, only one layer of the hidden layer can meet the needs of use. When the hidden layer reaches three layers, it can reflect the mapping of any continuous function. The BP network model structure is shown in Fig. 1.
BP neural network is a kind of algorithm learning process with teacher guidance. Learning mainly has the following steps: (1) initialize setting weights w and θ. It is clear that the connection weight of all neurons between the input layer and the hidden layer is wij. The neuron connection weight from the hidden layer to the output layer is wjk. The hidden layer threshold is designated θj. It is assumed that the threshold θk of the output layer neuron is a small value in the range [0,1]. (2) Determine the input vector xi = (x1, x2, ⋯, xm) of the input value. Calculate the matching expected output vector $$\overset{\wedge }{Y_i}=\left(\overset{\wedge }{Y_1},\overset{\wedge }{Y_2},\cdots, \overset{\wedge }{Y_n}\right)$$. The value of the xi input vector is input to the neuron node of the input layer. According to the conditional formula $${x}_j^i=f\left(\sum \limits_{i=0}^n{W}_{ij}{x}_i-{\theta}_j\right)\kern0.5em \left(j=1,2,\cdots, u\right)$$, the positive-oriented flow calculation is performed, or the counter-oriented flow calculation is performed according to $${y}_k=f\left(\sum \limits_{k=0}^n{V}_{jk}{x}_j-{\theta}_k\right)\kern0.5em \left(k=1,2,\cdots, n\right)$$. (3) Calculate the error between the output layer neuron output value and the expected output value. If the error result meets expectations, the training is completed. If the difference is too large, it will enter the reverse calculation of the model calculation again. Finally, after repeated calculations of the correction function, the weights satisfying the requirements are obtained, the calculation of the model ends, and the signal output is performed. The BP neural network learning algorithm flow is shown in Fig. 2.

### 3.2 BP neural network optimization

By improving the parameter values of the BP neural network, the defects that may be encountered in neural network learning can be overcome. The optimization method for the network is to adjust the maximum weight value to accommodate the decrease of the error. In the network backpropagation stage, the theoretical weight value and the previous weight value are partially superimposed as the actual weight value in this study. After the last time the weight changes, you can use the momentum factor to express the effect of the change and use which momentum factor to adjust the algorithm based on the change. When the momentum factor mc is equal to zero, the weights are adjusted according to the gradient descent method. When mc = 1, the adjustment of the new weight value will be the same as the change value of the last weight, which will exclude the adjustment brought by the gradient descent method. After adding the momentum item into the calculation, the network weight value is in the smoother range at the bottom of the error surface. ∇f(w(n)) will become very small, so w(n + 1) ≈ w(n). In order to prevent w(n + 1) = 0 from appearing and help the network to jump out of the local minimum, the weight adjustment formula is optimized as formula (1).
$$\nabla w\left(n+1\right)={m}_c\left[w(n)-w\left(n-1\right)\right]-\left(1-{m}_c\right)\eta \nabla f\left(w(n)\right)$$
(1)
Among them, n is the training times, mc is the momentum coefficient, ηB is the learning rate, and mc ∈ [0, 1] is set. When the error accuracy is assumed to be 0.0008, an extra-kinetic energy method is used to calculate the non-fitting linear function, the learning rate is set at random, and the average of 30 calculations is taken. Thus, the curve between the kinetic energy coefficient and the learning time is obtained, as shown in Fig. 3. It can be seen in the figure that after the introduction of the momentum item, the speed of network learning has been improved. This adjustment method can effectively avoid the occurrence of local minimum values of the network and reduce the recurrence of errors.
The learning process of neural networks is the key to solving the problems of fixed learning rate and frequent adjustment of weights. In order to overcome effectively the learning rate is too small, the algorithm convergence speed is too slow, the learning rate cannot reach the goal, the learning rate formula is optimized to help the network to reduce the correction, and to avoid the weight value exceeds the optimal value of the gradient. The learning rate optimization formula is shown in formula (2).
$${\displaystyle \begin{array}{l}\eta (n)=\left\{\begin{array}{c}a\times \eta \left(n-1\right)\\ {}a\times \eta \left(n-1\right)\end{array}\right\}\begin{array}{c}\kern0.5em E(n)<E\left(n-1\right)\\ {}\begin{array}{cc}& E(n)>c\times E\left(n-1\right)\end{array}\end{array}\\ {}\\ {}w(n)=w\left(n-1\right)-\eta (n)\times \frac{\partial E(n)}{\partial W(n)}\end{array}}$$
(2)
Here, η(n) is the learning rate at N iterations, and E(n) and E(n + 1) are the values of the two previous error functions. The constants a = (2,1), b = (1,0), c = [1,1.2]. When E(n) < E(n + 1), it means that the error is decreasing. This is because the learning rate will increase to a time of the past, and the convergence speed will increase. When E(n) < E(n + 1), it indicates that the error is increasing. At this time, the weights are in an over-adjusted state in the iteration, and it is necessary to reduce the learning rate to the original b times in order to avoid crossing this gradient direction, resulting in local optimal weights. In order to reduce the error of the BP neural network in training, the value of the directivity function indicating the search direction can be continuously reduced. This situation will cause the network convergence speed to drop, and the results of local minima search will not meet the need. A conjugate gradient algorithm is introduced at this time to provide a direction vector for the search. This algorithm uses the error function of the weight setting range as a function of the secondary line, and can be calculated at a time to obtain an accurate approximation. The implementation process is to first set the objective function = min E(w)   w ∈ R. When the minimum value of the error function is searched in the gradient direction, formula (3) can be obtained and the network can be corrected accordingly.
$${\displaystyle \begin{array}{l}E\left(w(n)+{\eta}_nd(n)\right)=\min E\left(w\left(n+1\right)\right)\\ {}w\left(n+1\right)=w(n)+{\eta}_nd(n)\\ {}d(n)=-g(n)+{\beta}_nd\left(n-1\right),d(0)=-g(o)\end{array}}$$
(3)
In order to solve the application of BP neural network, since it is not easy to determine the network structure, and it can easily lead to local minimum and other deficiencies, genetic algorithm is introduced to optimize it and improve the performance of the neural network. Genetic algorithm has a great advantage in global search. Based on the population, it uses individual fitness as a criterion to evaluate subsequent heritage operations. Not only is the global search capability good, but in the presence of mutation operators, the local search ability is also improved. The main form of genetic algorithm optimization for neural networks is: the first is to optimize the topological structure between the various layers in the neural network and various parameters of the neuron. In order to solve the problem that the hidden layer and the number of nodes in the neural network cannot be accurately determined, the genetic algorithm is used to optimize the topology structure before optimizing the network parameters. If the neural network structure is clear, use genetic algorithms to update the neural network thresholds and weights. And the number of neural nodes in the audit network, the number of hidden layers in the network is calculated.
The algorithm steps for optimizing the structure of the BP neural network using the genetic algorithm are as follows: first, the encoding and decoding are preprocessed. After determining the number of hidden layers and the number of nodes in the network, randomly generated chromosomes are decoded to form a corresponding neural network. The neural network is configured with an initial weight of 1 for learning and training. The difference between the expected network output value and the actual output value for each code is calculated, and the individuals with the smallest error are set as the first-generation parent. The above steps are repeated until the best individual is found. The individual’s corresponding network is the optimal neural network structure. The combination of BP neural network and genetic algorithm can effectively improve the speed of convergence, reduce errors and improve accuracy, and can effectively reduce the situation into the local minimum, so that the algorithm model gives a more scientific and optimal prediction results. The algorithm flow chart of the genetic algorithm introduced BP neural network is shown in Fig. 4.

## 4 Result analysis and discussion

In order to verify the research on the land evaluation of farmland transfer based on BP neural network proposed in this paper, the land transfer related data of M province from 2010 to 2017 was selected to conduct simulation experiments. The collected sample data is normalized before the model is built. The maximum and minimum normalization method is used here. Through this method, the data is transformed linearly and the data is mapped to the 0–1 range. From the neural network’s characteristic of acquiring knowledge through learning, the collected sample data is divided into test sets and training sets. The database from 2010 to 2015 was used as training data. The data from 2016 to 2017 is used as test data to evaluate the accuracy of the neural network. The MATLAB toolbox is used. The experimental process includes three parts: neural model design, neural network training, and evaluation network simulation. The model design is first performed. The relevant parameter design is the transfer function of the hidden layer and the output layer uses the tansig function, the training function is the purelin function, the real interval is set to 12, the network learning rate is 0.002, and the maximum training frequency is 230 times. The target error is 0.75*10(− 10). In order to avoid overfitting in the experiment, the network model’s prediction ability is reduced and the generalization ability is not good. Three times cross validation are used, that is, all the prediction results are the average values after cross-validation.
The parameter determination process of the BP network is a process of repeated learning accumulation. In the BP network model, the effect of a separate independent variable on the dependent variable depends not only on the size of this independent variable but also on the values of other independent variables. Therefore, in the experiment, the trained network was used to predict the 2016 and 2017 data. The experimental results are shown in Table 1 below. From the data in the table, it can be seen that the relative error is less than 1%, which proves that the neural network evaluation and prediction system constructed in this paper has a very good accuracy. Figure 5 shows the variation of the BP neural network error.
Table 1
Comparison between predicted and actual values
Particular year
Predicted value (%)
Real value (%)
Relative error
Managers’ proportion
The proportion of researchers
Managers’ proportion
The proportion of researchers
Managers’ proportion
The proportion of researchers
2016
3.35
62.21
3.31
61.79
0.56
0.04
2017
3.22
64.34
3.42
64.05
0.61
0.52
Figure 6 shows the BP neural network model constructed in this paper. The total number of experiment iterations was 69, and the best number of iterations was 63. At the 61st time, the best cross-validation was MSE = 0.22171.

## 5 Conclusion

The BP neural network has many advantages such as qualitative accuracy, high efficiency, and strong ability to deal with nonlinear problems. As an important representative of high-efficiency mathematical models, this model has made positive contributions to the management of all aspects of human society. Therefore, the use of BP neural network mathematical model is proposed to study land value evaluation of rural land circulation. After an in-depth analysis of the BP neural network structure, randomly determining weights and thresholds for the network will result in overfitting. Human factors affect the number of nodes, and it will lead to network learning time is too long, the number of iterations is too much, and the learning rate is not strong. Other targeted optimization is conducted to improve the accuracy and efficiency of the BP neural network prediction model. The weights and other parameters are improved, and the structure of the BP neural network is optimized. The network learning rate formula is improved to help the network reduce the amount of correction and overcome the slow convergence of the algorithm. A genetic algorithm is introduced to determine the network structure. After improving the performance of the neural network and other improvements, the optimized BP neural network is simulated. From the verification results, the construction of the evaluation model based on BP neural network is successful and can lay a good foundation for the promotion of rural revitalization. However, there are still some areas that can be improved in this study. The next step can be to conduct in-depth research on improving the prediction accuracy of neural models.

## Funding

The study was supported by “Hunan Natural Science Foundation Project (Grant No. 2018JJ2368),” “Hunan provincial decision consultation project (Grant No. 16JCC055),” and “Hunan provincial Education Department Project (Grant No. 15A174&16C1465)”.

### Author’s information

A Z, Master of Business Administration, Associate professor, Graduated from Guangxi University in 2007, working in Shaoyang University. His research interests include regional economics and rural land issues. Z L, Doctor of management, Professor, Graduated from Central South University in 2014, working in Shaoyang University. Her research interests include regional economics and rural land issues.

### Competing interest

The author declare that they have no competing interests.

### Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Unsere Produktempfehlungen

### Premium-Abo der Gesellschaft für Informatik

Sie erhalten uneingeschränkten Vollzugriff auf alle acht Fachgebiete von Springer Professional und damit auf über 45.000 Fachbücher und ca. 300 Fachzeitschriften.

### Springer Professional "Wirtschaft"

Online-Abonnement

Mit Springer Professional "Wirtschaft" erhalten Sie Zugriff auf:

• über 58.000 Bücher
• über 300 Zeitschriften

aus folgenden Fachgebieten:

• Bauwesen + Immobilien
• Business IT + Informatik
• Finance + Banking
• Management + Führung
• Marketing + Vertrieb
• Versicherung + Risiko

Testen Sie jetzt 30 Tage kostenlos.

### Springer Professional "Technik"

Online-Abonnement

Mit Springer Professional "Technik" erhalten Sie Zugriff auf:

• über 50.000 Bücher
• über 380 Zeitschriften

aus folgenden Fachgebieten:

• Automobil + Motoren
• Bauwesen + Immobilien
• Business IT + Informatik
• Elektrotechnik + Elektronik
• Energie + Umwelt
• Maschinenbau + Werkstoffe​​​​​​​

Testen Sie jetzt 30 Tage kostenlos.

Weitere Produktempfehlungen anzeigen
Literatur
Über diesen Artikel

Zur Ausgabe