Microsoft Word - cet-01.docx CHEMICAL ENGINEERING TRANSACTIONS VOL. 46, 2015 A publication of The Italian Association of Chemical Engineering Online at www.aidic.it/cet Guest Editors: Peiyu Ren, Yancang Li, Huiping Song Copyright © 2015, AIDIC Servizi S.r.l., ISBN 978-88-95608-37-2; ISSN 2283-9216 A Simulate Prediction and Analysis of Jilin Province Rural Tourism Development Based on BP Neural Network Li Li School of culture and tourism, Jilin Province Economic Management Cadre College, 130012, China p0wer316@sina.com The BP neural network is the most mature and widely used artificial neural network model. It has the simple structure, strong operability and good self learning ability. This paper first introduce the neural network in general. This paper choose the BP neural network as a method of simulated prediction of 2014 to 2023 Jilin Province rural tourism revenue and scenic area. Then this paper identify the prediction indicators of the neural network, complete the network training. The results shows the BP neural network has a good simulation ability of arbitrary complex nonlinear system and a strong fitting ability. The simulation process also indicates the BP neural network has many deficiencies, including difficulty in initial learning rate selection and low convergence speed. Keywords: Pearman correlation method, Rural tourism, BP Neural Network 1. Introduction Neural network, also known as artificial Neural network is composed of a large number of neurons in a certain topology of the interconnection network, it can reflect the basic characteristic of the human brain, it is the abstract and simplify simulation of the human brain. Neural network involves neural science, physics, mathematics, statistics and computer science, etc. The neural network can study the intelligent behavior through simulate the human brain’s physiologic structure and function. In recent years, although there have been many kinds of artificial neural network (ANN) model have been proposed and studied in depth, but consider for all the existing neural network model, back propagation (BP) learning algorithm neural network model or its improved version have a proportion of 80% to 90%, Yin, F. et al(2011) reported. Theoretically, three layer BP neural network can simulate any complex nonlinear system as long as it has enough hidden layer nodes. Therefore, the BP neural network is widely used in pattern recognition, signal processing, automatic control and forecasting. The application of BP neural network have to determine the network structure in advance, and the network structure optimization has always been the hot spot for researchers around the world, Cui Dongwen (2013) reported. BP neural network structure includes the number of input layer neurons, the number of output layer neurons, the number of hidden layer neurons. The key of Network structure design lies in the design of the hidden layer structure, specifically refers to the number of hidden layer and number of neurons in hidden layer. The number of the hidden layer determine the network's memory capacity, generalization ability, training speed and quality of the output response. A small number of hidden layer will produce non-convergence learning, low recognition ability and low generalization ability of the network. Since the reform and opening up, China’s national economy has achieved rapid development, people's living standard and life quality have greatly improve, from now on the economic development of our country goes into the stage of industrialization, the non-agricultural industries become the leading industry of the national economy. The traditional agricultural industry development falls behind in recent years, it become a bottleneck of our country’s society's development, Boskovic, T. et al(2013) and Komppula, R. (2014) reported . With the guidance of national policies, driven by the local governments, rural tourism is the revitalization of regional economic and effective means to increase farmers' income, the rural tourism just nourish in a short span of 20 years in the vigorous and rapid development of our country. 2. Modeling Method The topology structure of the BP neural network is shown in figure 1. DOI: 10.3303/CET1546104 Please cite this article as: Li L., 2015, A simulate prediction and analysis of jilin province rural tourism development based on bp neural network, Chemical Engineering Transactions, 46, 619-624 DOI:10.3303/CET1546104 619 Figure 1. The topology structure of the neural netwok 3. Prediction model of Jilin province rural tourism industry development 3.1 Prediction indicators’ raw data Rural tourism as a new type of tourism, it’s still in the development in Jilin province. The statistics about rural tourism first began at 2006. This paper select the historical data from year 2009 to 2013, shown in table 1 and figure 2. The data all come from the provincial statistical yearbook, other national economic and social development statistic comes from the provincial bulletin and government report. Table 1 The factors affecting the rural tourism income and scenic area indicator Year 2009 2010 2011 2012 2013 GDP (billion) 271.9 324.1 397.5 475.3 573.4 Tertiary Industry production growth (billion) 82.8 95.4 209.8 246.8 269.1 Fiscal revenue (billion) 18.6 36.0 42.4 23.6 54.0 Fiscal spending (billion) 21.9 36.1 42.8 32.5 63.2 Investment in Fixed Assets (billion) 104.7 143.5 190.6 250.0 325.7 Agriculture, forestry, water conservancy and environmental expenditure (billion) 0.9 1.1 1.4 1.8 2.0 Urban disposable incomes (Yuan) 14006 15411 17797 20208 22275 The per capita net income of rural areas (Yuan) 5266 6237 7510 8598 9621 Reception capacity of scenic area (Million hectares ) 52 85 100 450 550 Scenic area (Million hectares ) 0.8 1.2 1.6 2.1 1.9 Number of scenic area 70 75 86 98 124 Number of tourism farmhouse 1560 2200 2780 2840 3210 Urban road total length (kilometer) 1480 1842 2115 2297 2482 Number of private cars 7928 75121 75623 76431 78145 Rural garden area (Million hectares ) 6542 8670 10999 12453 15362 Figure 2. The factors affecting the rural tourism income and scenic area 3.2 The prediction indicators Consider the large number of rural tourism income and rural tourism scenic area indicators, only a small number of indicators are included in this paper. The large number of indicators will increase the workload, not able to find the main factors, it is a necessary procedure to screen the indicators Fernandes, Teixeira, et al (2012) reported. Due to the indicators’ different dimension, this paper use a normalized processing to get the unified dimensionless data, and then with the Spearman correlation analysis method to analyze each 620 indicator’s correlation to the rural tourism income and rural tourism scenic area, then we get the main indicator of the prediction model. The result is shown in table 2 and 3. From table 2, the GDP, Tertiary industry production growth, Investment in Fixed Assets, Agriculture, forestry, water conservancy and environmental expenditure, The per capita net income of rural areas, Reception capacity of scenic area, Number of scenic area, the 8 indicators has a close correlation to the rural tourism income with correlation coefficient greater than 0.8. From table 3, the GDP, Tertiary industry production growth, fiscal spending, rural garden area, the 5 indicators have a close correlation to the rural tourism scenic area. Table 2. The correlation of rural tourism income and all the indicators indicator Correlation indicator Correlation GDP X1 0.901 The per capita net income of rural areas X8 0.925 Tertiary Industry production growth X2 0.845 Reception capacity of scenic area X9 0.891 Fiscal revenue X3 0.253 Scenic area X10 0.429 Fiscal spending X4 0.359 Number of scenic area X11 0.865 Investment in Fixed Assets X5 0.854 Number of tourism farmhouse X12 0.359 Agriculture, forestry, water conservancy and environmental expenditure X6 0.923 Urban road total length X13 0.341 Urban disposable incomes X7 0.865 Number of private cars X14 0.329 Table 3. The correlation of rural tourism scenic area and all the indicators indicator Correlation indicator Correlation GDP X1 0.865 Fiscal spending X6 0.864 Primary Industry production growth X2 0.342 cultivated area X7 0.429 Tertiary Industry production growth X3 0.794 Agriculture, forestry, water conservancy and environmental expenditure X8 0.386 Reception capacity of scenic area X4 0.298 Rural garden area X9 0.684 Fiscal revenue X5 0.214 Rural tourism revenue X10 0.337 3.3 Simulation Prediction Model Based on the BP neural network modeling analysis, the relationship of rural tourism income and impact indicators related to rural tourism income is: ( ) = = ++= 3 1 6 1 21 j i jijij BBPWVT (1) In (1): T - rural tourism income target output P - rural tourism income indicator actual input or rural tourism scenic area indicator actual input V - the weights of the network layer W - the weights of input layer B1 - input layer threshold 4. Network training and simulation prediction 4.1 Network Training After the BP neural network is generated and initialized, we can use the existing "input - target" sample vector data for the network training. (1) Data Pre-processing. In order to improve the efficiency of neural network training, use premnmx function on "input - target" sample to standardize the data as pre-processing, which will make the data falls into interval [-1,1]. Assume SX as the standardized influential indicator on rural tourism income and rural tourism scenic area, SY as the standardized rural tourism income and rural tourism scenic area, as shown in table 4 and 5. 621 Table 4. Standardized Rural Tourism Income Data Year Input P Output T SX1 SX2 SX5 SX6 SX7 SX8 SX9 SX11 SY 2009 -0.7168 -0.8492 -0.8082 -0.9987 1.0000 -0.3017 -0.9805 -0.9864 -1.0000 2010 -0.7560 -0.8751 -0.8143 -0.9787 0.6214 -0.4332 -0.9865 -0.9912 -0.7725 2011 -0.9432 -0.9716 -0.9511 -0.9995 -0.6147 -0.8623 -0.9963 -0.9987 0.4240 2012 -0.9266 -0.9603 -0.9317 -0.9986 -0.4878 -0.8216 -0.9879 -0.9967 0.4340 2013 -0.9156 -0.9621 -0.9153 -0.9989 -0.4264 -0.8001 -0.9859 -0.9967 1.0000 Table 5. Standardized Rural Tourism Scenic Area Data Year Input P Output T SX1 SX3 SX6 SX9 SY 2009 -0.7487 -0.6801 1.0000 -1.0000 -1.0000 2010 -0.7800 -0.6691 1.0000 -1.0000 -0.5823 2011 -0.8004 -0.6536 1.0000 -1.0000 -0.2352 2012 -0.7644 -0.5987 1.0000 -1.0000 0.3863 2013 -0.7799 -0.5760 1.0000 -1.0000 1.0000 (2) Network Training. With the newff function of Matlab, we can create a forward BP neural network. This paper will set up a three layers BP neural network. After the processing of sample data, the BP neural network can be trained with Traingdm algorithm as follow: ),,(, TPnettraintrnet = (2) In (2), the P is Input sample vector set, T is the corresponding target sample vector set, net on each side of the equal sign is the neural network object before and after training, tr stores the erros and logs of the training process. Network simulation. The simulation is achieved through sim function with the trained neural network, the sim function is: ),( PnetsimA = (3) In (3), P is the input sample vector set, T is the corresponding target sample vector set, net on each side of the equal sign is the neural network object before and after training, tr stores the erros and logs of the training process, A is the simulation result. The simulation error degree is shown in figure 3 and figure 4. Figure 3. Rural Tourism Income Neural Network Simulation Error Degree The figure 3 of rural tourism income network training error curves shows that the network iteration 807 times to complete the training, and error reach the minimal error target. Therefore, the simulation effect is good, the neural network model is acceptable. The figure 4 of rural tourism scenic area network training error curves shows that the network iteration 3715 times to complete the training, and error reach the minimal error target. Therefore, the simulation effect is good, the neural network model is acceptable. 622 Figure 4. Rural Tourism Scenic Area Neural Network Simulation Error Degree 4.2 The Prediction Result With the model and trained neural network in this paper, the simulation of rural tourism income and scenic area in the following 10 years can be predicted, as shown in table 6,7 and figure 5. Table 6. The prediction of 2014 to 2023 rural tourism income Year 2014 2015 2016 2017 2018 Predicted Income ( Billion) 1.03 1.39 1.6 1.73 1.84 Year 2019 2020 2021 2022 2023 Predicted Income (Billion) 2.07 2.15 2.37 2.51 2.63 Table 7. The prediction of 2014 to 2023 rural tourism scenic area Year 2014 2015 2016 2017 2018 Predicted Area (Million hectares ) 2.4 2.8 3.2 3.7 4.1 Year 2019 2020 2021 2022 2023 Predicted Area (Million hectares ) 4.9 5.6 6.5 7.4 8.6 Figure 5. The 2014 to 2023 rural tourism revenue and scenic area 5. Conclusions This paper use rural tourism income and rural tourism scenic area as prediction objects. Based on 2009 to 2013 rural tourism data, this paper builds a simulation prediction model and predicts the future 10 years rural tourism income and scenic area. The prediction results shows that the rural tourism industry in Jilin province is rapidly developing, and in the year of 2014 to 2023, the rural tourism industry will continue growth greatly. The simulation process also indicates the BP neural network has many deficiencies, including difficulty in initial learning rate selection and low convergence rate. The fixed and variable vector algorithm in BP neural network have a very low convergence speed, which is shown in this paper, the initial training take 3715 epochs. Thus the variable vector algorithm in BP neural network need to be improved to reduce the initial learning epochs. And also the BP neural network has a poor predict extrapolation effect. When processing of 623 a small interval training, the forecast period and the training period tends to be inconsistent, when processing of a large interval training, the predict extrapolation effect will get worse. To improve the BP neural network in future research, the improvement can focus on the time series algorithm and unipolar Sigmoid function algorithm References Bai Shizhen, Li Shuang. (2013). Study on financial risk assessment of Supply Chain Based on BP neural network. Commercial research, 55 (1), 27-31.DIO:10.3969/j.issn.1001-148X.2013.01.005. Cui Dongwen (2013). Based on the improved BP neural network model of water resources vulnerability assessment in Wenshan, Yunnan, the Yangtze River academy of Sciences, 30 (3), DIO:10.3969/j.issn.1001- 5485.2013.03.001. 1-7. Fons, M. V. S., Fierro, J. A. M., & y Patiño, M. G. (2011). Rural tourism: A sustainable alternative. Applied Energy, 88(2), 551-557. DIO:10.1016/j.tourman.2013.07.007. Guresen, E., Kayakutlu, G., & Daim, T. U. (2011). Using artificial neural network models in stock market index prediction. Expert Systems with Applications, 38(8), 10389-10397. DIO:10.1016/j.eswa.2011.02.068. Gao Shutao (2013). Cs algorithm to optimize BP neural network for short-term traffic flow forecasting. Computer engineering and Applications (9), 106-109.DIO:10.3778/j.issn.1002-8331.1212-0188. Komppula, R. (2014). The role of individual entrepreneurs in the development of competitiveness for a rural tourism destination–A case study. Tourism Management, 40, 361-371. DIO:10.1016/j.tourman.2013.07.007. Nair, V., Munikrishnan, U. T., Rajaratnam, S. D., & King, N. (2015). Redefining rural tourism in Malaysia: A conceptual perspective. Asia Pacific Journal of Tourism Research, 20(3), 314- 337.DIO:10.1080/10941665.2014.889026. Park, D. B., Lee, K. W., Choi, H. S., & Yoon, Y. (2012). Factors influencing social capital in rural tourism communities in South Korea. Tourism Management, 33(6), 1511-1520. DIO:10.1016/j.tourman.2012.02.005. Randelli, F., Romei, P., & Tortora, M. (2014). An evolutionary approach to the study of rural tourism: The case of Tuscany. Land Use Policy, 38, 276-281. DIO:10.1016/j.landusepol.2013.11.009. San Martín, H., & Herrero, Á. (2012). Influence of the user’s psychological factors on the online purchase intention in rural tourism: Integrating innovativeness to the UTAUT framework. Tourism Management, 33(2), 341-350. DIO:10.1016/j.tourman.2011.04.003. Santos, C. N., & Thomaz, R. C. C. (2014). Tourism at rural areas of the municipality Rosana/SP: possibilities of inclusion. Turismo y Desarrollo: Revista de Investigación en Turisme y Desarrollo Local, 7(16). Su, B. (2011). Rural tourism in China. Tourism Management, 32(6), 1438-1441. DIO:10.1016/j.tourman.2010.12.005. Wang Zeyu (2013). Based on the improvement of BP neural network of real estate bubble measure evaluation study. Financial theory and practice, 34 (4), 95-98.DIO:10.3969/j.issn.1003-7217.2013.04.019. Yin, F., Mao, H., Hua, L., Guo, W., & Shu, M. (2011). Back Propagation neural network modeling for warpage prediction and optimization of plastic products during injection molding. Materials & design, 32(4), 1844-1850. DIO:10.1016/j.matdes.2010.12.022. 624