Abstract
The world has been overturned by the power of the Internet, and all walks of life are constantly changing with the trend of the times. Therefore, with the help of the effective technology and means of the Internet, people can start from a more complex perspective. Analyze the social environment and economic situation, and focus on poverty. By actively monitoring the number of poor people in society, we can grasp the actual situation of poverty. Starting from the depths of the problem, aiming at the characteristics of poor groups, this paper excavates the scale data of poor people. Finally, the improved BP neural network model is used to predict the time when the society gets rid of poverty and reduces poverty. The results show the following: (1) After value judgment and value test, the experiment determined 12 indicators and proportion of poverty. The most prominent indicators of contribution rate are T2, T4, and T5. The most significant changes in test results were T8 and T12. (2) The reformed BP network has less iteration times, shorter training time, and higher prediction accuracy. The network converges 30 times, and the fitting accuracy reaches the best, and the ideal state is 0.000936578. Training the poverty alleviation data from 2016 to 2020 with the model, the final poverty alleviation work has achieved remarkable results. (3) According to the hidden layer nodes of neural network structure, the final number of hidden layer nodes is determined to be 11. (4) Comparing the absolute error and APE error of the prediction results. The predicted value is very close to the actual value, which can correctly reflect the mapping relationship of poverty reduction and the result is reliable. (5) Compare the predicted values of BPNN model before and after modification; it is found that the new BPNN model is more accurate in prediction. The predicted population is close to the actual population, and the error level curve approaches 0%. After empirical test, the improved BPNN model is accurate and effective. Through the poverty reduction prediction of the model, people can observe the actual situation more intuitively. By evaluating the effectiveness of social poverty alleviation, we can optimize the next poverty alleviation strategy and adjust the strategic policy.
1. Introduction
The prosperity and strength of the country are closely related to poverty reduction. With the continuous development of the global economy, the gap between the rich and the poor is getting bigger and bigger. The problem of poverty is gradually revealed, which seriously hinders the development of society. Consolidate the strength of society, accurately predict poverty reduction, and formulate a series of strong and effective assistance policies to make social development more stable and harmonious. This makes the precise poverty alleviation work measures of far-reaching significance. For prediction problems, the computer technology of neural network is very mature and has obvious advantages and achievements. The modeling effect is stable and simple, and the prediction accuracy and efficiency are good. People use a series of improved models to put into the market and can make actual predictions of various scenarios. The following literature provides many excellent ideas, which can provide reference for the research topic of this paper.
A comprehensive model of double-layer BPNN and preceding section state is proposed, and the prediction accuracy of bus arrival time is improved by 57.25% compared with that of traditional BPNN model [1]. Improved artificial bee colony algorithm built a model based on MABC-BPNN and predicted the ultralow energy consumption of grassland residents in western Inner Mongolia [2]. To compare the fitting and prediction effects of the BPNN and SARIMA models for the incidence of Class B infectious diseases [3]. Using BPNN learning prediction model and PID feedback [4]. In order to predict the concentration of PM2.5, a BPNN multifactor prediction model based on genetic algorithm was proposed [5]. BPNN optimized by cuckoo search algorithm can predict PMV index of aircraft cabin [6]. An improved immune genetic algorithm and BP neural network model is proposed to predict steel sales [7]. A genetic neural network model is established to predict river water quality [8]. Lasso regression model was introduced to screen the influencing factors, and combined with BP neural network model, cucumber short-term price forecast was carried out [9]. In order to improve the accuracy of hyperspectral estimation of water content of oil peony seeds, the water characteristic absorption parameters and BP neural network were used to design the model [10]. An improved BP neural network combined with factor analysis was used to design a model to predict the duration of traffic accidents [11]. The Drosophila optimization algorithm is introduced, and the BP neural network model is improved to predict the time of getting rid of poverty [12]. Analyze and discuss the framework and mechanism of sustainable poverty reduction effect and accurate monitoring of poverty alleviation effect of poverty alleviation model [13]. Taking the sustainable livelihood system of farmers as the analytical framework, this paper discusses the reduction of poverty population in social areas [14]. Based on Alkire and Foster, we put forward three critical values to measure China’s multidimensional poverty reduction effect from poverty alleviation and returning to poverty [15].
There are a lot of research work under the background of Internet. At present, there are many improvements and applications of BPNN model in the market. According to different needs of different industries, the above literature improves BP neural network model for scenario prediction. This paper will study and discuss the characteristics of social poverty groups, analyze the influence of various factors on poverty alleviation, and then build a reasonable poverty reduction prediction model. After prediction and observation, people can better formulate corresponding poverty alleviation suggestions according to the social poverty dynamics and avoid problems such as poverty alleviation rebound. This way can better comprehensively observe the future trend of poverty alleviation. Governments and departments can make corresponding decisions and policies more pertinently.
2. Theoretical Basis
2.1. Overview of Poverty Reduction
Poverty has been an enduring hot topic since ancient times. Strive to eliminate poverty, promote various working concepts on poverty reduction, and implement every policy indicator. The emergence of precise poverty alleviation has achieved remarkable results in promoting poverty reduction. Eliminating absolute poverty is a historic milestone event. China has made outstanding contributions to global poverty reduction, contributing more than 90%. How to reduce poverty and let people live and work in peace and contentment is a problem that every country is seriously trying to eliminate. Poverty alleviation and difficulties tackling is related to the great plan of human survival and development. For this paper, we divide the poor groups into two groups at present. First, absolute poverty; under social conditions, this part of the group is affected by various factors, and it is difficult for them to use their legal income to meet the most basic personal and family living conditions. This group is our top priority in poverty reduction. The second is relative poverty; this group can maintain basic survival needs. However, compared with the average living standard of society, they still have a big gap, with little income level and poor living conditions. This part of the group is the most difficult to eliminate and will exist for a long time. We need to fight against relative poverty for a long time, so it is necessary to carry out the design work of predicting poverty reduction. In 1976, the academic circles conducted a large-scale relative poverty survey through OECD. It is pointed out that 50% of the average income of a country or region is determined as the relative poverty line of the country or region. This line is gradually accepted internationally, which is referred to as the international poverty line for short. Some countries or regions set 60% of the average income for marking. In this paper, considering the actual situation of our country, we carefully consider using this method to draw lines. The international poverty line is raised by 10%, which is determined as the standard for dividing relative poverty in this paper. Calculate the poor population. Share [16]:
Calculate the incidence of poverty. Where represents the incidence of poverty in different regions; represents the size of poverty population in each region; and refers to the total population size.
There are many definitions of poverty. So far, there is no concept that has been fully recognized by the whole world. Especially, the academic research on poverty in China is late, and there is no mature theoretical system yet. Therefore, in order to help understand the origin of poverty, this paper refers to some foreign authoritative experts and scholars and gives a certain source and explanation. By referring to the definition of poverty, we can have a deeper understanding of the emergence and development of poverty, so as to discuss the core issues of poverty reduction from the root, as shown in Table 1.
2.2. Data Mining
Data mining [22]. is a commonly used prediction algorithm. It can find valuable information hidden in it by analyzing massive data and data mining algorithm. This process is closely related to computer science. All walks of life can analyze and mine information from different angles, so as to help adjust the strategy and make better analysis, evolution and decision analysis.
Only by ensuring that the whole mining process meets the requirements can we get the goals we need. For most data mining processes, we can briefly summarize them. The first is to clarify the goal and define the problem, and the specific model is determined by the actual problem. Then, the problem-related data mining database is established to collect and integrate the data. After processing the data, analyze the data, and then build and implement the mining model. We highly refine the concrete performance of its basic prototype design. As shown in Figure 1.
In this paper, association rules mining algorithm is introduced to find various relationships in large databases, so as to find the implicit relationship between items. Because Apriori algorithm in many association rules algorithm, it is the most widely used, the most mature. Therefore, we use the core recursive algorithm to find frequent itemsets through candidate itemsets. Then, the credibility and confidence are screened out centrally, and the final association rules are obtained. Several important formulas:
Formula (3) is related to support; Formula (4) is related to confidence, which can determine the frequency of and ; Formula (5) is the ratio of credibility to expected credibility, which is called promotion degree, where contains the case probability of item .
2.3. BP Neural Network
BPNN [23].can also be used in prediction studies. Its full name is BP neural network model. This is a sigmoid function composed of input layer, hidden layer, and output layer. After the information propagates forward, if the actual output is different from the expected one, reverse error propagation will occur. It is mainly aimed at single-layer network structure, in order to solve the linear separable problem. Modify Widrow-Hoff [24], ADALINE [25]. The infrastructure of BPNN is simple, as shown in Figure 2.
Input to the algorithm:
Adjust network parameters. Mean square error function:
For approximate mean square error, its steepest descent algorithm:
Definition of neuron sensitivity:
Among them, the input of the network is represented by , and the target output is represented by ; is the vector of network weights and offset values; denotes the learning speed, and denotes the m-th layer of the neural network.
Re-represent the steepest descent algorithm:
Then, find the sensitivity of the last layer of the network:
Recursive relation of sensitivity:
2.4. Time Series
In the era of big data, the importance of data need not be emphasized. However, data is very dependent on time, which is a hot issue. Time series and its related forecasting work are commonly used methods in our daily life. As the name implies, we sort a set of data we need in a certain time interval order. This time depends on the situation and can be set as a time unit such as year, month, day, and hour according to needs. In this way, our data has the attribute of time, which makes it more intuitive and convenient to analyze the data situation. Because of the limitations of regression prediction, it is difficult to determine the rules or relationships among its other variables. By analyzing time series, we can find some phenomena or laws that are not easily perceived directly. These data can well analyze and predict the indicators to be analyzed. Therefore, we use time series to participate in the model design of BP neural network and carry out a series of adjustment and optimization work. Simple linear relationship between time series and time.
Trend of quadratic curve:
3. Improved BPNN Predictive Poverty Reduction Model
3.1. Construction of Influencing Factor Index
There are many influencing factors for poverty. If we want to build a quantitative prediction model for poverty reduction, we need to know what factors can determine poverty. To set up indicators, we should not only intercept a single indicator. Instead, according to the relevant economic data, we should summarize and deal with it, and comprehensively consider the problem of constructing indicators. We introduce a SPCA model to analyze the poverty situation in China. According to the environment and disaster situation and five important impact causes, the real causes of poverty can be found, which further affects specific poverty alleviation policies and various strategies. As shown in Figure 3:
Finally, it was evaluated and agreed by experts and scholars. In this paper, the specific setting of 12 indicators is obtained, as shown in Table 2.
Because of these different indicators, the statistical units of their original data are different. As a result, we cannot directly use these data for comparison. Therefore, if we want to compare data, we must first standardize the data. -Score method is used to unify the caliber, change every original data, and carry out data preprocessing for analysis and comparison, where stands for standard deviation; stands for the smallest value in the sequence, Namely,
3.2. Improved BP Neural Network
Although BPNN model is often used, it has many defects and deficiencies. First of all, BP neural network is affected by the algorithm setting, so it is difficult to get a small and suitable learning rate. Especially because of the steepest descent method, its convergence speed is slow and the training time is very long, which is difficult to meet the demand. In addition, BP network has a complex network structure, and it is easy to fall into the problem of local extremum.
In this paper, time series is introduced to facilitate the analysis of poverty reduction effect. In addition, the improved BPNN model also uses LMBP algorithm for standard numerical optimization. Here, suppose the column vector:
The Levenberg-Marquardt algorithm is a variant of Newton’s method. At the same time, it is also the learning rule of the network:
The performance function:
Its gradient is
Approximate representation of the Hessian matrix:
Suppose is small:
Gauss-Newton method for obtaining deformation:
Find the Jacobian matrix. Assumption error vector:
Finally, the recurrence relationship between sensitivity augmented matrices:
3.3. Model Structure Design
According to the analysis, MATLAB platform is used for modeling, a simple BP neural network. Input 1, carry out neural network processing and output corresponding contents. As shown in Figure 4.
Training network. Input layer-hidden layer, its weight distribution structure model: . Input 1, through weight assignment, the weight dot product function is obtained; converge the function into multiple signals and output the final target, as shown in Figure 5.
4. Experimental Analysis
4.1. Development Environment
In this study, we carry out design analysis. The framework of its prediction model is mainly based on Framework. We use C # language for development. A specific software and hardware development environment is necessary, as shown in Table 3.
4.2. Proportion of Impact Indicators
Although we finally set 12 indicators in this paper, it involves quantitative data. Only after quantification will our indicators be more accurate and standard. Finally, judge whether the prediction result is accurate or not. The value was used to screen the data, and the significance was used to determine the limit, as shown in Table 4.
Set NUM to 50 and 100 and pass test. 1000 pieces of data were extracted to participate in quantitative analysis. Because the data with value greater than 0.05 in the judgment table is not significant, we must eliminate this part of the data index. From the figure, we can find that the values of 12 indexes all meet the requirements of this paper. Therefore, all 12 indicators are up to standard. However, the test results and contribution rates of each index are different. These two indicators fluctuate widely. Among them, the most prominent data of index contribution rate are T2, T4, and T5. In the test results, the data increase is the most significant, and they are T8 and T12, as shown in Figure 6.
4.3. Model Training and Verification
4.3.1. Training Model
Combining with predecessors’ experience and practice, we train the model. The Bayesian regularization method is mainly used. The iteration of BP network after modification is better. The prediction accuracy is higher, and the training time of the model is also short. After 30 trainings, the network converges. The fitting accuracy reaches the best performance; the ideal state is 0.000936578.
Then, the absolute poor people from 2016 to 2020 were selected for training test. Compare the actual poverty alleviation situation with the poverty alleviation situation predicted by the model. The number of data sets will be modeled with selected training samples. We can find that the training value should be wirelessly close to the best value, and the final iteration effect is the best. The ratio of test set to training set is set to 3: 7. By analyzing the data in the chart, we can find that the model of experimental training and testing in this paper is very effective. The number of people who get rid of poverty and reduce poverty gradually decreases with the increase of years. This aspect illustrates the remarkable achievements in poverty alleviation. Social poverty reduction work is good and has achieved results, as shown in Figure 7.
4.3.2. Hidden Layer Analysis
The hidden layer in neural network model is analyzed. There are three nodes, called Node 1, Node 2, and Node 3. The number of each node increases in turn. Their training time increases with the increase of the number of nodes. However, the absolute mean square error is different for different nodes. Among them, the error value of Node 2 is the smallest. Finally, the neural network structure with 11 hidden layer nodes is selected, as shown in Figure 8.
4.3.3. Absolute Percentage Analysis
For the prediction results, an absolute percentage error is needed. This error, called APE for short, exists between the exact value and the approximate value. If the difference is smaller, it proves that the predicted value is close to the actual value. This shows that the accuracy of the prediction results is better, as shown in Figures 9 and 10.
We still use the improved BP neural network to predict the number of people who will reduce poverty from 2016 to 2020. The results are used as experimental data. Combining the two charts, we can find that the absolute values of error and APE in 2020 are on the high side. It can be known that the prediction accuracy in 2020 is obviously lower than that in other years. Overall, the accuracy of the model is between 80% and 90%. The accuracy of the model is high, which can correctly reflect the mapping relationship of poverty reduction, and the results are reliable.
4.4. Predictive Analysis
In order to better highlight the effect of the improved model, the original BP model is specially added. The prediction results of the original traditional algorithm are compared. At the same time, it is necessary to compare the relative errors, as shown in Figure 11.
The traditional algorithm test fluctuates greatly, the error curve is ups and downs, very unstable, and the error situation is more and more large. However, after observing the improved BPNN prediction results, it is found that there is little difference between the predicted number of people and the actual population. The total error level curve approaches 0%.
5. Conclusion
Using data mining technology, through the powerful functions of the Internet, based on time series and LMBP network, and relying on BPNN model, a poverty reduction prediction model is constructed. The effectiveness of poverty alleviation can be evaluated by the data of prediction model. The research results of this article show the following: (1)Firstly, the proportion of each index is determined by experiment. Through value judgment and value test, 12 indicators affecting poverty are selected. The data with the most prominent contribution rate are T2, T4, and T5. The most significant changes in test results are T8 and T12(2)Secondly, the reformed BP network has less iteration times, shorter training time, and higher prediction accuracy. The network convergence only needs 30 times of training, which can make the fitting accuracy reach the best performance, and the ideal state is 0.000936578. The poverty alleviation data from 2016 to 2020 are used for model training(3)According to the neural network model, the hidden layer is experimented. The number of hidden layer nodes is determined to be 11(4)Comparing the absolute error and APE error of the prediction results. The accuracy of the model is basically between 80% and 90%. The predicted value is very close to the actual value, which can correctly reflect the mapping relationship of poverty reduction and the result is reliable(5)Finally, compare the number of people out of poverty from 2011 to 2020. By predicting the predicted values before and after the transformation of BPNN model, the data are compared. It is clear that the newly designed BPNN model is more accurate and superior to the traditional algorithm. There is little difference between the number of people predicted by the new model and the actual population, and the final total error level curve approaches 0%
In this paper, compared with the traditional BP neural network, the improved BPNN prediction model makes up for the shortcomings of the original. Under the condition of more stable network, the need of monitoring poverty can be fully met. The prediction effect of the model is good and has higher prediction accuracy. Although our model has met the performance requirements of prediction, it is a qualified auxiliary tool. The accuracy of the prediction model should be more accurate; the article can use cluster analysis to divide the poor population, so that the division of poverty degree is more detailed, thus making the assistance work more accurate; because the improved BPNN model has too many parameters, the calculation is complex and difficult to realize; a series of index policies should be constructed to influence the poor. These works need more theoretical and research support for further optimization and improvement.
Data Availability
The experimental data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declared that they have no conflicts of interest regarding this work.
Acknowledgments
This study was supported by the National Social Science Foundation of China: Research on the innovation system of Chinese decentralization for financial return to its origin (19XJL005).