Fund Project: Shanghai University Knowledge Service Pinghe "Shanghai Business Service Knowledge Service Center" Construction Project (ZF1226); Shanghai "Engineering Shanghai Business School" Business Administration "Key Discipline Construction Project Graduate Tutor, Research Interests: Data Mining, Financial Engineering , Management Science and Engineering.
The location of the hotel is based on a comprehensive consideration of many influencing factors, to determine the best operating location of the hotel's fixed service facilities, in order to meet the needs of consumers for short-term accommodation, whether it is directly related to the future success of the hotel business, It is an important basis for hotel managers to develop their business strategies. Site selection is a very important task in the pre-planning demonstration of the hotel. The exploration of hotel location problems at home and abroad began in the 1980s. At first, qualitative analysis was the main method. Recently, the (fuzzy) multi-criteria decision-making method and hierarchy were adopted. Analytic Hierarchy Process (AHP) and Neural Network (NN) technology, through the establishment of evaluation index system and location model for quantitative analysis, have achieved certain research results, but for multi-index decision-making problem of hotel location These research methods still have certain deficiencies in the modeling principle. For example, (fuzzy) multi-criteria decision-making method and AHP need to hire experts to determine the weight of evaluation indicators artificially. The modeling process is more complicated. When the number of indicators is large, it is difficult for experts to make weights for indicators. Reasonable judgment, the evaluation results have a certain subjective randomness. The NN model is suitable for the modeling of nonlinear problems, but it needs a sufficient number of training samples. It is a very difficult technology to master. It needs artificial trial and error to determine the structure of the network. Therefore, when the modeler is not comprehensive on the NN technology, When understanding, it is easy to establish a "pseudo-model," which affects the validity of the evaluation results. If the network structure is determined to be too large, the number of network connection weights is much larger than the number of training samples, and there is no training process. Using the test sample to monitor the training process, the phenomenon of “overtraining†and “overfitting†will inevitably occur during the training process. Strictly speaking, the NN model thus established does not have any generalization ability, and the evaluation result is also invalid. In this paper, the NN model is re-established in the sample data in this paper to verify the unreliability of the evaluation results. At the same time, the applicability and feasibility of the NN model and PPC model in the study of microsite location of hotel location are analyzed.
PPC technology is suitable for the modeling of nonlinear high-dimensional data. In view of the multi-indicator impact of hotel micro-location problem, this paper uses PPC technology to establish the hotel's micro-location model, and establishes the evaluation through the optimization of the established projection index function. The best projection vector of the index, and can be integrated into a one-dimensional projection value to achieve an objective and comprehensive evaluation of the sample. Zheng Longsheng has a unique research on the location of the hotel. The empirical study of establishing the PCC model with Zheng Longsheng's sample data shows that the method can not only effectively describe the merits and demerits of the selected samples, but also determine the evaluation index through the optimal projection vector value. The nature of the analysis and ranking of the importance of indicators provides an important basis for hotel investors and managers to develop business strategies.
When the larger the index is, the better the index is, the better. When Xi is the normalized data of the sample, maxX and minX are the maximum and minimum values ​​of the first indicator. It should be specially pointed out that the method of normalization of extreme values ​​does not affect the nature of indicators. PPC technology can be used both as exploratory research and as deterministic analysis, combined with normalized methods and optimal projection vectors. The sign of the value can determine the nature of the evaluation index. When the larger the better the naturalization mode is adopted, if the optimal projection vector value is greater than 0, the indicator is a positive indicator, otherwise it is a negative indicator. ) Representation in one-dimensional space after mapping: At present, the projection index function is often constructed by the product of the standard deviation of the sample projection value and the local density, ie Q(a)=SzDz, where Sz and Dz respectively represent the sample projection value Z(i The standard deviation and local density, the expressions of the two are: (4) where E(Z) is the mean of the sample projection value Z(i); r(i,k)=|z(i)-z(k )| represents the distance between two samples in one dimensional space; R is the local density window radius, u(Rr(i,k)) is a unit step function, and the function value is equal to 0. The mapping of the projection vector, sample points The distribution characteristics in the dimensional space should be consistent with the statistical principle that the sample points in the local projection point group are as dense as possible, and the projection points on the whole are scattered as much as possible. Therefore, the optimal projection vector can be determined by solving the maximum value of the projection index function. : Substituting the optimal projection vector a obtained in step 3 into the function formula (3), obtaining the comprehensive projection value of each sample, and evaluating and sorting the samples according to the magnitude of the Z(i) value. The value Z(i) is more The larger the sample, the better.
Second, the evaluation index system of micro-site selection of business hotels and the source location of sample data are important manifestations of the implementation of consumer-centered business enterprises, and are the key factors for the success or failure of business strategy. Therefore, it is necessary to comprehensively consider the influence of many factors. . The micro-site selection of business hotels is part of the commercial site selection. It is also a complex multi-index evaluation system due to factors such as industrial agglomeration, land cost, social environment, business environment, transportation factors, expansion factors, and urban planning factors. Starting from these seven factors, Zheng Longsheng selects the relevant industry correlation X1, the number of competitors X2, the distance X3 between competitors, the land cost X4, the public facilities perfection degree X5, the public order quality X6, the surrounding business commerce X7, Distance to central business district X8, distance to traffic station X9, convenience to traffic station XW, number of parking lots X, convenience of city X12, convenience to business activity area XB, size of expandable space XM, A total of 16 indicators of the coordination degree XB of the existing buildings and the degree of coordination with the future urban construction planning were established to establish a comprehensive evaluation index system for the micro-site selection of business hotels, and the value of each evaluation index was quantified by a 5-point semantic scale method.
Table 1 The optimal projection vector of the three groups of samples Considering the focus of this paper is to compare the applicability, feasibility and effectiveness of the PPC technology and NN technology in the study of hotel micro-site selection, so we will not repeat how to select To evaluate the indicators, we directly quoted Zheng Longsheng's evaluation index system and 58 sets of sample data. In the empirical modeling research analysis, the PPC business hotel micro-location model is established by 55 training samples, and the optimal projection vector of the evaluation index is obtained. Then, one of the three candidate samples is calculated by using the projection vector. The dimensional projection value is based on the projection value and then analyzes the advantages and disadvantages of the three alternative address samples.
Third, empirical modeling of micro-site selection of business hotels 1. Establishment of micro-location model of business hotel based on PPC The above 55 sample data are preprocessed according to the larger and better normalization method, and then brought into MALTAB simulation software simulation. Computation, multi-agent genetic algorithm is used to optimize the projection vector. The parameters of the algorithm are set as: the size of the agent grid Lsize=20, the competition probability Po=0.2, the crossover probability Pc=0.1, the mutation probability Pm=0.1, the function to be optimized The number of variables variablenum=16, the maximum number of loop iterations is set to 300, and the optimal projection vector of the calculated index is shown in Table 1. In addition, in order to verify the accuracy of the calculation results, this paper adds three virtual indicators Xl7, Xi8, and Xi9 to the normalized sample data, and specifies that the values ​​of the three virtual indicators are x17=1, x18=x1, X19=1-x1, that is, x19 is the standard value after the pre-processing of the index m with the smaller the better. The sample data after adding the virtual indicator is brought back to the simulation model (in this case, the number of variables of the function to be optimized is variablenum=19) is calculated and calculated: the projection vector value of xi and the three virtual indicators ((, (, (calculated result) The projection vector value of the indicator in the four indicators is equal to 0, the index x18 is equal to the projection vector value of the index x1, and the index x19 is opposite to the projection vector value of the index x1, which indicates that the multi-agent genetic algorithm is optimized in the modeling process. The parameter setting is reasonable, and the calculation result of the PPC-based business hotel micro-location evaluation model is reliable, and the projection vector at this time is the optimal projection vector.
The best projection vector is obtained accurately and stably. PPC technology exudes human factors interference when determining the weight of indicators. The determination of its weight depends entirely on the sample data itself, which is typical of “data-drivenâ€.
model. Therefore, in order to study the influence of sample size on the modeling results, this paper extracts two sets of sample data from the original sample, and calculates the optimal projection vector of the two sets of sample data to compare with the original 55 samples to verify the modeling. The stability of the results. In the process of sample extraction, according to the requirement that the number of PPC modeling samples is more than the number of evaluation indicators, the number of evaluation indicators is 1.5 times 25 samples and the number of evaluation indicators is 2.5 times 40 samples. The two sets of sample data are preprocessed according to the same larger and better normalization method, and then respectively brought into the simulation model calculation, and the optimal projection vector is obtained as shown in Table 1.
Comparing the optimal projection vectors of the three sets of samples in Table 1, it can be seen that the optimal projection vectors of the 40 sample sets and the 55 sample sets are completely consistent, while the 25 sample sets are in the indicators x3, x5, x9, xW, xU. The optimal projection vector value differs from the presence of 40 sample sets and 55 sample sets. In order to further explore the difference between the best projection vector values ​​of these five indicators, the previous research in this paper analyzes the statistical characteristics of the indicators, selects the standard deviation of the mean and dispersion degree of the concentrated trend, and the skewness and distribution tip that characterizes the degree of skew. The statistical analysis results of the 25 indicators of the erratic table are based on the 55 training samples to obtain the best projection vector of the evaluation index, and then analyze the advantages and disadvantages of the candidate samples. Therefore, the key to the PPC modeling is the degree. The kurtosis 4 data statistics indicators, trapped in space This article only lists the statistical analysis results of these five indicators, as shown in Table 2.
The statistical analysis results of the five indicators in Table 2 can be concluded. Compared with the 40 sample groups and the 55 sample groups, the standard deviation of the index X3 in the 25 sample groups is small, the skewness is large, and the kurtosis is obvious. If the ground is too large, the skewness of index X5 is small, the skewness of index X9 is obviously small, the kurtosis is obviously larger, the skewness of index Xi is obviously smaller, and the kurtosis of index x is obviously larger. It is shown that the data structure characteristics of these five indicators in the 25 sample groups are different from those of the latter two groups, which may explain the reasons for the different modeling results. The above empirical modeling studies show that when there are enough typical samples, the number of samples is generally required to be more than the number of evaluation indicators. The PPC technique can be used to obtain a robust optimal projection vector; When the sample has similar data structure features, PPC technology modeling can also obtain the best projection vector that matches. Zheng Longsheng selected 58 hotels in the same city as research samples, and used NN technology to model and predict the advantages and disadvantages of three candidate samples for 55 training samples. The same paper used PPC technology to model 55. The sample obtains the best projection vector of the evaluation index, and calculates the projection values ​​of the three candidate samples, and analyzes the advantages and disadvantages of the three candidate samples.
2. Evaluation results and analysis of the candidate site samples The optimal projection vector and the candidate address samples are obtained by normalizing the standard values ​​according to the larger the better (see Table 3) and substituting the formula to obtain 3 alternatives. The projection values ​​of the address samples are: Pl=-0.3001, P2=0.6681, P3=0.1750. The larger the projection value is, the better the sample is. It is obvious that the projection value of P2 is obviously larger than P3 and P1 is the best place for site selection.
Table 3 Standard data after normalization of the three candidate samples First, the nature and weight analysis of the evaluation indicators. The integrated projection value is essentially a linear combination of the evaluation indexes with the best projection vector as the coefficient. From the magnitude of the optimal projection vector value and whether it is greater than 0, the nature of the evaluation index and the importance of the analysis index can be determined. The larger the projection value, the better the sample, so the best projection vector value is greater than 0 is the forward index, and the best projection vector value is less than 0 is the negative index. Analysis of the results of 55 sample groups in Table 1 shows that the optimal projection vector values ​​of the indicators xi, X2, X4, X5, X6, X7, X8, Xi3 are greater than 0, and the positive optimal projection vector value is less than 0, which is negative. To the indicator. The larger the absolute value of the optimal projection vector value is, the larger the corresponding index is. The higher the contribution rate to the projection value is. The absolute value of the best projection vector of the 16 indicators is analyzed. The importance of the index is X12 and X1. The absolute value of the best indicator vector is greater than 0.3 is a relatively important indicator, the sum of their contribution rate to the integrated projection value is 55%, so hotel investors and managers should pay special attention to these when selecting hotel sites. Indicators, from these indicators to analyze the geographical characteristics of the hotel, and develop the hotel's future business strategy.
Second, the correlation analysis between the evaluation indicators. The above analysis shows that the optimal projection vector value of 8 indicators is less than 0 is a negative index. This paper analyzes the correlation between indicators by taking the weighted indicators X3 and xXM as examples. First, there is a significant negative correlation between the distance X3 between the index competitors and the industry correlation XI around the positive indicator, the number of competitors X2, and the distance X8 from the central business district. The correlation coefficients are -0.573, -0.549, - 0.386. Secondly, there is a negative correlation between the city's traffic convenience degree xu and the industry correlation Xi around the positive indicator, the surrounding business prosperity X7 and the distance X8 from the central business district. The correlation coefficient is -0.291, -0.246. -0.283. The final indicator expandable space size x14 has a significant negative correlation with the industry correlation XI around the positive indicator, the public facility perfection degree X5, the surrounding business prosperity X7 and the distance X8 of the indicator from the central business area. Correlation coefficient The order is -0.367, -0.394, -0.511, -0.366. Therefore, there is a negative correlation between the above indicators and the positive indicators with large weights, and it is reasonable for the model to judge that they are negative indicators.
The distance X3 between the competitors of the indicators is a positive indicator when analyzed from the theory of industrial agglomeration. When analyzing the competition between the same industry, it should be a negative indicator. Therefore, it is difficult to judge the nature of the indicator; the indicator of the city's traffic convenience X12 The subjective judgment of XM and expandable space size should be positive indicators, but there is a significant negative correlation between them and the positive indicators. Therefore, the modeling results in this paper show that the negative indicators are reasonable. When there are a large number of indicators, there will inevitably be correlations between indicators. This paper uses PPC technology modeling to determine the weight and nature of the evaluation indicators based on the data structure characteristics of the samples themselves, avoiding the influence of human subjective factors and ensuring The objectivity and effectiveness of the evaluation results.
3. Problems in Zheng Longsheng's thesis and the impact on the modeling results First of all, Zheng Longsheng's training sample results are obtained according to the expert scoring method, and this is the final evaluation result we need, and the linear relationship between the evaluation results and the evaluation indicators. The NN model is a typical data-driven model “garbageingarbageoutâ€. Modeling the above samples with NN technology does not change the linear nature between the evaluation results and the evaluation indicators. Therefore, it is not necessary to use NN technology to establish a complex and nonlinear. model.
Secondly, Zheng Longsheng adopted a too large network structure, and did not use the test sample to monitor the training process. During the training process, the phenomenon of “overtraining†is likely to occur. The generalization ability of the model is not verified, and the evaluation result is also invalid. Since Zheng Longsheng did not list the results of the sample evaluation, after contacting the author, the other party's computer failed and failed to obtain the result data. However, the one-dimensional projection value of the evaluation result of PPC is linear with the evaluation index, and the evaluation result of the expert scoring method is the same in form. Therefore, it is feasible to perform simulation calculation as the result of the training sample. Since Zheng Longsheng's training target error is 0.00001, and the actual training sample error at the end of training is 0.000079, this paper understands its target error as 0.0001. The PPC evaluation results and 55 samples are brought into Statsoft's neural network. The software re-establishes the NN model and finds that when the linear evaluation model is established using the network structure without hidden layer, the training accuracy is easy to reach the target error of 0.6692, 0.2134) and the result of PPC is basically the same; and when the number of hidden nodes is 32. At the time, 20% of the total sample size is selected as the test sample and the test sample. Since the number of network connection weights at this time is 577, which is much larger than the number of training samples, when the training sample error reaches 0.0015, the training process has already occurred. Training phenomenon, as shown. Continue training until the training sample error satisfies the requirement of less than 0.0001. Table 4 shows the output of the three candidate samples in the three modeling.
From the results in Table 4, it can be seen that the modeling results of the table 4 NN model of the three candidate samples are hidden layer number of training samples. The results of the PPC modeling results mostly deviate from the expected values, and the rankings are also good or bad. All have changed, P2 may be optimal, or it may be the worst, that is, each modeling output is different. Therefore, Zheng Longsheng's modeling results have two obvious errors: First, the nonlinear NN model is used to model a linear system. The actual modeling study in this paper shows that this has no practical meaning. When using a linear model without hidden layers. The training sample error can also easily meet the target error requirement. That is, when the results of obtaining samples using linear models such as AHP, expert scoring, and PCA are brought into NN technology modeling, this is wrong in itself, and when a simple linear model can solve the problem, it is generally not built using NN technology. Complex nonlinear model. The second is to determine the network structure is too large, and the test sample monitoring training process is not used. The actual modeling research shows that the model is prone to “overtraining†in the training process, and the evaluation results are very unstable. significance. NN technology has been widely used in many fields for its parallel computing and adaptability to nonlinear problems, and has achieved good research results. However, when researchers blindly model without fully understanding NN technology, they often Not getting the desired result.
This paper uses PPC technology to establish a comprehensive evaluation model for micro-location of business hotels. Firstly, the optimal projection vector of the evaluation index is obtained with 55 samples, and the optimal projection vector is verified by adding dummy variables and studying the influence of different sample quantities. Then, the one-dimensional projection values ​​of the three candidate samples are calculated by using the optimal projection vector, and the three candidate samples are ranked according to the projection value. The example modeling research shows that the method is feasible.
The PPC model determines the optimal projection vector based on the structural characteristics of the sample data itself, avoiding the interference of human factors. The weight and nature of the evaluation index can be clearly analyzed by the optimal projection vector value. The indicators that are subjectively difficult to judge for the distance between competitors and the two indicators that are subjectively judged as the positive indicators for the convenience of the city and the size of the expandable space, this paper analyzes the correlation with other positive indicators. The three indicators of sexual judgment are negative indicators, which ensure the accuracy of the evaluation indicators and the weight determination, and ensure the objectivity and effectiveness of the evaluation results.
As a complex network structure, the NN model is mainly suitable for the evaluation of nonlinear problems, but has strict requirements on the quantity and quality of training samples. If the evaluation result has a linear relationship with the evaluation index itself, then there is no practical significance in using NN technology modeling; when the model uses too large a network, and does not use the test sample to monitor the training process, it is likely to appear during the training process. Without any application value, the evaluation results are also invalid. Therefore, for the number of training samples is small, and the sample results rely on the micro-location problem of the hotel obtained by expert scoring, it is not suitable for modeling with NN technology.
Wood Brow Pencil,Wooden Brow Pencil,Matte Eyebrow Pen,Slim Eyebrow Pencil
Guangzhou Bijiang Cosmetics Co., Ltd. , https://www.bijiangcosmetics.com