A Prediction Method for Postconstruction Settlement of Pile-Soil Composite Subgrade Based on Fuzzy Comprehensive Evaluation

&is paper presents a postconstruction settlement prediction method for pile-soil composite subgrade based on the multilevel fuzzy comprehensive evaluation principle. In this method, the variation range of postconstruction settlement can be obtained from a simple calculation based on the basic data of actual engineering. Firstly, according to the characteristics of influencing factors in the construction of soft soil subgrade, the evaluation index set and two-level factor index sets were selected. &e grading standards of the evaluation index and factor index were determined according to the allowable value of the standard and the numerical simulation results. Secondly, each factor index was standardized, and the normal distribution function in the form of exponential was used to construct the standard membership function for the first and second factor indexes. Finally, the comprehensive evaluation matrix of postconstruction settlement of composite subgrade was constructed based on the entropy weight method. &e variation range of postconstruction settlement was predicted by the principle of maximummembership. &e example analysis shows that the predicted results of the prediction method and the field measurement method are in good agreement, indicating that the proposed method can realize the postconstruction settlement prediction of composite subgrade, and the results are more accurate and more instructive.


Introduction
e postconstruction settlement of the composite subgrade of expressways has always been concerned by both academia and industry, especially in some special sections, such as bridge head transition sections and the sections containing structures [1,2]. Higher requirements are put forward for the postconstruction settlement of subgrade in these sections. So people often adopt the foundation treatment method of pile-soil composite subgrade. Compared with other foundation treatment methods, pile-soil composite subgrade has certain advantages in improving the stability of subgrade and reducing settlement. However, there are still some large postconstruction settlements in the treatment section of pile-soil composite subgrade, such as car jump at the bridge head. Many scholars and engineers have carried out much useful work in the prediction of postconstruction settlements by means of theoretical analysis [3], numerical simulation [4], and on-site observation data fitting [5,6]. Previous related works on postconstruction settlement prediction are discussed and tabulated in Table 1.
In this paper, based on a large amount of postconstruction settlement field test data, a new prediction method of postconstruction settlement of composite subgrade is proposed using the multilevel fuzzy comprehensive evaluation method.
is method calculates the variation interval of postconstruction settlement through a simple calculation by only considering the basic data from actual projects. e calculation results are beneficial to the optimization of design and improvement of construction quality. In the establishment process of this method, it is difficult to calculate weight and membership degree accurately, and there are many factors involved. Hence, the entropy weight method, standardized membership function, and two-level evaluation system are introduced. In this way, not only the calculation accuracy is improved but also the index of influencing factors involved in the comprehensive evaluation in each level is reduced, which is more convenient for practical applications.

Need for Research
In recent years, artificial intelligence technology has been developed rapidly, but at present, artificial intelligence technology still has some shortcomings and limitations [16][17][18]. Firstly, it needs more data and data support to build the model. When the data are insufficient, the error is often large. Secondly, sometimes, in the modeling process, some engineering features or states need to be represented by numbers, which will lead to the loss of information. Finally, the results obtained by the artificial intelligence technology model are usually qualitative indicators, while the quantitative indicators are relatively few and not rigorous enough. However, in addition to the aforementioned disadvantages, artificial intelligence technology also has obvious advantages [19][20][21]. First of all, it is an important tool for system analysis and will not cut the relationship between influencing factors and results. Secondly, it can solve complex relationships between multiple factors and predict unknown data. Finally, its analysis results are often more concise, easy-to-make decisions.
Considering that the advantages of artificial intelligence technology have a great help to solve complex engineering problems, some artificial intelligence technologies have been widely used in civil engineering. To predict the settlement of rock-socketed piles, Danial et al. [22,23] presented the development of a hybrid ANN-based model named PSO-ANN (or neuro-swarm) with detailed modeling process and a new model based on gene expression programming (GEP). Li et al. [16] use least squares support vector machines (LSSVMs) based on multipoint measurement to monitor and predict foundation pit displacement. In addition, fuzzy comprehensive evaluation (FCE) method is also a typical artificial intelligence technology; it can better solve the fuzzy but difficult to quantify the problem and is suitable for all kinds of nondeterministic problems. e FCE method has been widely used in mineral [24], environmental [25], slope [26] and other engineering areas [27]. However, its application in postconstruction settlement of composite subgrade has not been reported. As the postconstruction settlement of composite subgrade involves many factors and has obvious randomness, it is meaningful to explore the method of uncertainty and establish a prediction model and a calculation method.

Evaluation Index and Influencing Factors
As the postconstruction settlement of composite subgrade can be affected by many factors, a two-level fuzzy comprehensive evaluation system considering the main influencing factors is selected. e evaluation system can make up for the shortcomings of the existing single-level evaluation methods and make the evaluation results more accurate and more instructive [20]. In the two-level fuzzy comprehensive evaluation system, one of our key works is to determine the reasonable evaluation index set and the factor index set of each layer for improving the accuracy and practicability of the prediction method.

Evaluation Index Set.
In the two-level fuzzy comprehensive evaluation system, an evaluation index v is constructed for describing the severity of the postconstruction settlement of the composite subgrade, and we have where V is a set that contains all the possible values of the evaluation index v. According to the requirements of the current specification in China, allowable values for the postconstruction settlements of high-quality highways at the bridgehead transition section, structural section, and general road section are 100, 200, and 300 mm, respectively. erefore, the element vm in the evaluation index set V is obtained based on the postconstruction settlement value. e smaller the postconstruction settlement value, the better the construction quality. In this paper, the evaluation index set contains four values to represents four quality grades (see Table 2. e space of the postconstruction settlement has been divided into four subsections, and the corresponding normalized classification intervals (w.r.t 400 mm) are also given in Table 1. For the postconstruction settlement larger than 400 mm, 400 mm is applied to the normalization process. Hence, we have a closed range of the postconstruction settlement after the normalization process.

Influencing Factors Set.
Since the postconstruction settlement of CFG pile composite subgrade-based road section involves many influencing factors, a two-level fuzzy comprehensive evaluation model is adopted to tackle the problem: 15 secondary influencing factors are constructed as the leaves of 4 primary influencing factors (geology, pile, cushion, and construction). Mathematically, we have where U i is the primary influencing factor, u ij is the secondary influencing factor, and n i is the number of secondary influencing factors included in U i .
e factor values of the secondary influencing factors vary with the engineering characteristics. Taking CFG pile composite subgrade in Guangdong Province as an example, this paper discusses the method of determining the secondary influencing factors. According to the collected sample data of a large number of CFG pile composite subgrade engineering projects, existing specification requirements, and on-site survey results collected by the authors, the range of the factor for each secondary influencing factor is provided as follows.
e influencing factors of the geology U 1 consist of compressed soil thickness u 11 , compression modulus u 12 , cohesion u 13 , and internal friction angle u 14 .
In practical projects, the depth of CFG piles should not exceed 30 m. Otherwise, the quality of composite subgrade is not easy to control. erefore, the range of compressed soil thickness u 11 should be 0 to 30 m. e cross-board shear strength of soft soil in Guangdong region is generally 4 to 12 kPa. In the shear strength index, the minimum cohesion c < 3.2 kPa, the minimum internal friction angle φ < 2.1°, the porosity ratio is generally greater than 1.0, and the compression coefficient is within 0.5 to 1.0 MPa −1 . erefore, the range of the cohesive force u 13 is 1 to 15 kPa, the range of the internal friction angle u 14 is 1°to 15°, and the range of compression modulus u 12 is 1 to 10 MPa. e influencing factors of the pile U 2 consist of pile modulus u 21 , pile length u 22 , pile diameter u 23 , and pile spacing u 24 .
Generally, CFG pile length range u 22 is 0 to 30 m, the pile diameter u 23 should be 0.35 to 0.60 m, and the pile spacing u 24 should be 1.4 to 3.0 m. As the CFG pile design strength range is 5 to 20 MPa, the rounding range in CFG pile compression modulus u 21 is ranging from 800 to 30,000 MPa. e influencing factors of the cushion U 3 consist of cushion thickness u 31 , cushion modulus u 32 , and cushion internal friction angle u 33 .
In practical projects, the general thickness of the CFG pile composite subgrade cushion is 0.3 to 0.6 m. Considering the action of the pile top support, the thickness of the cushion u 31 is in the range of 0 to 0.8 m. As the cushion is mainly medium coarse sand and gravel, the cushion modulus u 32 ranges from 50 to 250 MPa and the cushion internal friction angle u 33 ranges from 15°to 60°. e influencing factors of the construction U 4 consist of filling height u 41 , filling rate u 42 , preloading height u 43 , and preloading time u 44 . e filling height of expressways is mainly affected by terrain and routes. Except for a few hill areas, the filling height usually does not exceed 20 m (including the converted thickness of the pavement structure layer). erefore, the filling height u 41 ranges from 1 to 20 m. e embankment filling regular loading rate is 20 to 70 mm/d, the loading rate usually is less than 20 mm/d, and the loading rate is greater than 70 mm/d for fast loading. For the convenience of the analysis, the filling rate u 42 ranges from 5 to 100 mm/d. Surcharge preloading mainly includes under-preloading, equal-preloading, and over-preloading. Under normal circumstances, the overloading height does not exceed 3.0 m, and the preloading time does not exceed 1 year. erefore, the preloading height range is 0 to 3.0 m, and the preloading time range is 0 to 360 days.

Factor Grading Standard of Secondary Influencing Factors
By analysing a large number of engineering samples, the finite-element numerical software can be used to calculate the single-factor influence of the representative engineering example of composite subgrade. To determine the factor grading standard of all the secondary influencing factors, the influence of the individual factor was analysed separately by taking five to six values within its range.
Taking CFG pile composite subgrade as an example, Plaxis finite-element numerical software can be used to analyse the law of single-factor influence. e detailed simplified processing, calculation method, and calculation process, as well as reliability demonstration will not be discussed due to the limitation of space. e specific method to determine the grading standard of the secondary influencing factors is as follows.

Quantification of Influence of Individual Factors.
According to the FEM simulation results, three categories can be concluded to describe the relationship between the postconstruction settlement changes and the secondary influencing factors.
is first category is named as positive attribute factor u + ij . As the factor u + ij increases, the postconstruction settlement value increases and the evaluation grade decreases. According to the simulation studies, compressed soil thickness u 11 , pile spacing u 24 , filling height u 41 , and filling rate u 42 belong to this category. e second category is named as negative attribute factor u − ij . As the factor value increases, the postconstruction settlement value decreases and the evaluation grade increases. Compression modulus u 12 , cohesion u 13 , pile modulus u 21 , pile length u 22 , pile diameter u 23 , cushion thickness u 31 , cushion modulus u 32 , internal friction angle of cushion u 33 , preloading height u 32 , and preloading time u 33 belong to this category. e third category is named as zero attribute factor. As the factor value increases, the postconstruction settlement value does not change or changes little. e internal friction angle of subgrade soil u 14 belongs to this category.
For analysing the physical quantities with different units, a normalization process is implemented to both positive and negative attribute factors (zero-attribute factor is considered as the positive attribute factor in this paper). e normalization formulas for both positive and negative attribute factors are as follows: According to formula (3) According to Figures 1 to 4, the contribution of each secondary influencing factor can be obtained, which can be used as the reference for weight distribution in the fuzzy comprehensive evaluation process.
Based on Figure 1, the postconstruction settlement is very sensitive to the compressed soil thickness q 11 and the compression modulus q 12 . Compared with q 11 and q 12 , the changes of the cohesion q 13 have a small effect on the postconstruction settlement.
e changes of the internal friction angle of subgrade soil do not influence the postconstruction settlement.
Among the influencing factors of the pile, the postconstruction settlement is sensitive to the pile modulus q 21 and pile length q 22 . e pile diameter q 23 has the smallest influence on the postconstruction settlement in Figure 2.
In Figure 3, the changes of the internal friction angle q 33 and the thickness of the cushion q 31 have major effects on the postconstruction settlement. In comparison, the influence of the cushion modulus q 32 is relatively small.
Among the construction-influencing factors, the preloading height q 43 and preloading time q 44 have a major effect on the postconstruction settlement, and the effects of the filling height q 41 and filling rate q 42 are relatively small.

Grading Standard.
According to the simulation and analysis results shown in Figures 1 to 4, the linear interpolation method is used to determine the relationships between the normalized factors q ij and the postconstruction settlement and we have q ij � f ij (S p ). Where S p is the postconstruction settlement, f ij ( ) is the linear mapping from postconstruction settlement to normalized influencing factor q ij . Based on the derived linear relationships f ij ( ), the values of the normalized factors q ij can be calculated when S p equals 100, 200, and 300 mm. As a result, the corresponding u ij can be obtained, and the grading standard can be e change of friction angle in the subgrade soil does not affect the postconstruction settlement S p . erefore, the entire range of internal friction angle belongs to the grade v 1 .
Classification standard of all the secondary influencing factors is shown in Tables 3 and 4.

Membership Functions and Weights
Based on the sensitivity analysis in the previous section and the classification standard of influencing factors in Tables 2  and 3     function is used to describe the influence of each secondary influencing factor. e entropy weight method is used to determine the weights of the primary and secondary influencing factors.

Standardized Membership Functions.
Normal membership function: the commonly used membership functions in geotechnical engineering area are normal-type functions, triangular fuzzy functions, trapezoidal functions, and ridgetype functions [28].
e parameter membership function that in geotechnical engineering should adopt an exponential normal distribution function was suggested [29]. erefore, the membership function [30] of the structural influencing factor u ij on the evaluation index is as follows: where k is the evaluation index of u ij and a (k) ij and c (k) ij are constants, satisfying a (k) ij > 0 and c (k) ij > 0. Standardization of membership functions: to give a standardized membership function, the classification interval of the secondary influencing factor is linearly converted into a classification standard interval with the same size as the evaluation index. According to the level k to which u ij belongs, the positive and negative attribute factors have been normalized separately. After standardized treatment of influencing factors, the classification standards of each factor are consistent with the classification ranges of the postconstruction settlement classification standards in Table 1. a (k) ij and c (k) ij can be calculated using the upper and lower boundaries of the classification intervals of each factor classification standard. For the evaluation index level v 1 ∼ v 4 , the normal distribution membership function in the form of the normalized index is where y ij is the normalized value of the influencing factor; r (k) ij is the standard membership value of each influencing factor, and k � 1, 2, 3, and 4. e membership of each secondary influencing factor can be obtained using formulas (6) to (9). After obtaining the membership value of each secondary influencing factor, a matrix R i can be constructed:

Weights of Influencing Factors.
In the fuzzy comprehensive evaluation of multiple influencing factors, the weight is a measure of the influence of each influencing factor on the evaluation index. e method of determining the weight can generally be divided into subjective empowerment and objective weighting methods. Common subjective empowerment methods have Delphi method, analytic hierarchy process, and statistical method. Common objective weighting methods include entropy weight method, principal component analysis method, and mean square error method. As the subjective empowerment method relies on human factors, such as expert experience which may cause bias on the evaluation results, the entropy weight method in the objective weighting method is used to determine the weight of the influencing factors.
Entropy is a physical quantity that reflects the randomness and disorder of subjects. e weight of each influencing factor can be constructed based on the entropy of each influencing factor [31]. e steps for determining the weights of influencing factors by the entropy weight method are provided as follows [32]: (1) Collect the influencing factor samples from T projects. e projects are ordered and t is the index of the project.
(4) Calculate the entropy E ij of the secondary influencing factor u ij with where ξ � 1/ln(T) is a constant related to the number of samples. p ij,t has three requirements: e larger E ij is, the smaller the information of U i contained in u ij is and vice versa. (5) Calculate the difference coefficient d ij of each secondary influencing factor E ij by And then calculate the weight w ij of u ij w.r.t. U i . e expression of w ij is According to formula (15), a weight vector for the secondary influencing factors can be obtained, we have where all weight factors satisfy n i j�1 w ij � 1 naturally. (6) Use formula (16) to calculate the normalized value Y i,t of the primary influencing factor U i,t for project t, and we have (7) Use the entropy weight method similar to the process from Steps 2 to 5, the weight of the primary influencing factors can be calculated. According to formula (11), the ratio P i,t and entropy E i of the primary influencing factor U i,t are computed by when P i,t � 0, P i,t ln(P i,t ) � 0. e difference coefficient that defines for the primary influencing factor U i is Advances in Civil Engineering e weight of the primary influencing factors on the evaluation indicators is A weight vector of the primary influencing factors can be obtained: where all weight factors satisfy 4 i�1 w i � 1.

Postconstruction Settlement Assessment Method for Two-Level Fuzzy Comprehensive Evaluation
After introducing the evaluation index, the influencing factor grading standard, the standardized membership function, the weight of the influencing factors, and a twolevel fuzzy comprehensive evaluation method of the postconstruction settlement of the CFG pile composite subgrade are proposed in this section.

Evaluation Procedure.
(1) According to the postconstruction settlement requirements of the CFG pile composite subgrade in different road sections and the factors affecting the postconstruction settlement, the evaluation index evaluation set V, the primary influencing factor set U, and the secondary influencing factor set U i are determined. e secondary influencing factors are the inputs of this model. (2) Derive the relationship matrix R i for each primary influencing factors based on formula (10). (3) Based on formula (16), determine the comprehensive evaluation vector B i for the primary influencing factor U i using where " ∘ " is the sign of the composition operator; this paper takes the composition operator as the addition and multiplication operator M (g+), namely, (4) Construct the relationship matrix R for the primary influencing factors: (5) Calculate the evaluation vector B for the evaluation index by using where  Tables 5 and 6. According to the secondary weight factor values shown in Tables 5 and 6, the weights of the internal friction angle (2.5%) of the subgrade soil and the filling rate (9.3%) are relatively small, and they have little effect on their corresponding primary influencing factors. e weights of all other secondary influencing factors exceed 20% and smaller than 50%, which is satisfied with our claim that the postconstruction settlement of CFG pile composite subgrade depends on many factors. e weights of all the secondary influencing factors are in consistent with the sensitivity analysis results discussed in the previous section.
Similarly, the weights of the primary influencing factors can be computed. e results are shown in Table 7.
e data in Table 7 show that the weights of the influencing factors of cushion, pile, construction, and geology are reduced in descending order, meaning their influences on the postconstruction settlement are reduced in turn. For composite subgrade, the cushion plays a role in adjusting the stress ratio of the pile and soil and determines the structural form of the bearing system. e pile is the main bearing part for undertaking the weight of the composite subgrade. Hence, the weight of the geological influencing factor is the smallest (15.7%) among all the primary influencing factors. However, none of the primary influencing factors can be neglected. e calculated weights of the primary influencing factors also indicate that the proposed entropy weight method can be used for analysing the relationships between the primary influencing factors and the postconstruction settlements.

Model Inputs.
A bridge transition section in Guangdong Province has been selected to validate the proposed evaluation method. e target bridge transition section has CFG pile composite subgrade. e geological conditions and construction information are concluded as follows. e compressed soil thickness u 11 is 22.6 m, the averaged compression modulus u 12 is 6450 kPa, the averaged cohesion u 13 is 9.5 kPa, and the internal friction angle u 14 is 6.7°. e pile diameter u 23 is 0.4 m, the pile length u 22 is 25 m, the pile spacing u 24 is 2.2 m, and the pile modulus u 21 is 8.0 GPa. e piles have a plum-type arrangement. e gravel cushion thickness u 31 is 30 cm, the cushion modulus u 32 is 100 MPa, and the cushion internal friction angle u 33 is 35°. e filling height u 41 is 5.3 m, and the filling rate u 42 is 80 mm/d. Due to the tight schedule, the preloading process has not been applied in this road section. In this case, the preloading height u 43 and preloading time u 44 are 0. e index value and normalized value of secondary influencing factors are shown in Tables 8 and 9.

Postconstruction Settlement Prediction.
Based on the information of the target road section, the relationship matrix R i of each secondary influencing factors can be obtained. Following the process mentioned in the previous section, the relationship matrix R can be computed and we have According to the principle of the maximum membership, it can be seen from the matrix R that the level of geological influencing factors is excellent, the level of pileinfluencing factors and cushion-influencing factors are good, and the level of construction-influencing factors is bad. e analysis conclusion is consistent with the reality that no preloading was performed to the target road section.
According to equation (25), the comprehensive evaluation vector for the postconstruction settlement of CFG pile composite subgrade can be obtained as According to the principle of maximum membership, the evaluation result of the postconstruction settlement is good. Namely, the range of the postconstruction settlement is [100 mm, 200 mm). e construction of the target road section was completed and opened to traffic in July 2015. e observation frequency of the postconstruction settlement is shown in Table 10 and the collected postconstruction settlement data are shown in Figure 5.
According to the field observation data, the current settlement is 88.7 mm. Using the hyperbolic method, the final postconstruction settlement of the CFG pile composite subgrade after 15 years operation is 164.23 mm, which satisfies the calculation and evaluation results. Since the road section is a transitional section at the bridgehead, the allowable value of its postconstruction settlement should be less than 100 mm. Hence, certain remedial measures, such as   grouting, are necessary for this road section so as to meet the regulations.

Conclusion
In this paper, a two-level fuzzy comprehensive evaluation method is proposed for evaluating the postconstruction settlement of the CFG pile composite subgrade. Four primary influencing factors and 15 secondary factors are selected as the indexes to build the two-level evaluation system. e entropy weight method and normal membership function with the normalization process are used to describe the fuzzy relationship between the evaluation results and the influencing factors.
A real bridgehead road section example in Guangdong Province has been selected to validate the proposed fuzzy comprehensive evaluation method. e evaluation results of the proposed method are in consistent with the field test results. Hence, the proposed fuzzy comprehensive evaluation method can be used to evaluate the construction quality of the CFG pile composite subgrade.
e FCE method requires a large amount of data and is complicated to calculate. Moreover, the samples used in this paper to calculate the weight of influencing factors are mainly from Guangdong Province, China, which has certain limitations. Moving forward, efforts will be made to explore how to reduce the subjectivity in the determination of evaluation factors and weights and therefore to construct a more rational and adaptive FCE method.

Data Availability
e data used to support the findings of this study are presented in the tables and figures.

Disclosure
is work is an original one conducted at Xuzhou University of Technology. e work described has not been submitted elsewhere for publication, in whole or in part.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this paper.

Authors' Contributions
All the authors have contributed to and approved this paper.