Study of Risk Evaluation for Complex Projects under BIM and IPD Collaborative Pattern Based on Neighborhood Rough Sets

: Faced with the existing problems in the construction industry, project managers have been aware that traditional project delivery patterns cannot satisfy the current building industry's pursuit of high-quality economic benefits or cope with the increasing volume and complexity of modern buildings. Therefore, whole-process comprehensive management has become a new development trend for engineering projects, the traditional project delivery pattern should be changed, and early-stage communication, cooperation, and information sharing among project participants should be strengthened. As for the BIM and IPD collaborative pattern, the emphasis is laid on collaboration among project members, and the core idea lies in benefit and risk sharing and full consideration of human resources, commercial structure and engineering system of projects. The advent of IPD pattern has well solved problematic issues encountered in engineering projects, where risk problem has attracted the highest attention. Risk factors of IPD projects were reduced and screened out based on attribute reduction theory of rough sets, and primary risk factors influencing smooth implementation of IPD projects were obtained.


INTRODUCTION
With rapid economic development, all kinds of construction projects are presenting trends of scale expansion and technological complication and constructing parties, which are having higher requirements, and these changes have proposed more professional and higher requirements for consultation, design and construction in the whole engineering project construction process, and even for economic benefits of the building industry. Faced with the existing problems in the building industry, more and more experts and project managers have been aware that traditional project delivery patterns cannot satisfy the current building industry's pursuit of high-quality economic benefits or cope with the increasing volume and complexity of modern buildings, so wholeprocess comprehensive management has become a new development trend for engineering projects. A survey by the American Institute of Architects (AIA) in 2017 pointed out that [1], 83% of proprietors required to change the traditional project delivery patterns and strengthen earlystage communication, cooperation and information sharing among project participants, which realize overall project optimization to the greatest extent. The vigorous popularization of building information modeling (BIM) technology even poses a greater challenge to the traditional delivery patterns, and its implementation has boosted the reform of the building industry and elevated the production efficiency of the building industry and overall project benefits. Integrated project delivery (IPD) pattern was used in the oil drilling platform project in the North Sea in Britain at the end of the 1990s at the earliest, which achieved success in Australian and American projects. IPD pattern lays the emphasis on collaboration among project members, and its core idea lies in risk and benefit sharing and full consideration of commercial structure, human resources, and engineering system of the projects. Based on BIM technology and by reference to related theories of lean construction, projects can reach both local and global optimum. Among general transaction patterns, project participants aim at safeguarding their own benefits but not caring about benefits of counterparties and overall project benefits, thus obstructing communication among project members, increasing workload of design change and site visa of the projects and reducing project efficiency and benefits. The advent of IPD pattern has well solved difficulties encountered in engineering projects.

RESEARCH STATUS OF RISK SHARING OF COMPLEX PROJECTS UNDER BIM-BASED IPD PATTERN
Risk sharing is an essential path for risk control with the main work being: taking overall consideration of risk control abilities and risk bearing abilities of project participants and reasonably allocating risk factors existing in the projects to the most advantaged participants for risk management and control. Scholars from many countries have investigated risk sharing mainly from qualitative and quantitative angles and focuses on risk preference degree, quantification of influence level of risk factors and benefit distribution. As for qualitative studies of risk sharing, Veg pointed out that the core principle of risk sharing was to realize win-win between participants [2], but there was no fixed normal form for risk sharing, but instead, schemes should be formulated according to concrete situation of the project. As indicated by Frederick, the final goal of project risk sharing was to reduce the total project cost to the minimum [3]. Govan put forward allocating more risk factors to preference parties of risks and parties with strong risk bearing abilities, so as to ensure controllability of overall projects for risk factors [4].
Quantitative research methods of risk sharing mainly include fuzzy mathematical method, statistical analysis method and case study method. Directing at uncertainty and particularity of risks, Elbarkonky proposed a fuzzy emergency discriminant model based on expert investigation method (Delphi method) to evaluate probability of occurrence and influence level of project risks [5]. Nasirzadeh put forward a cooperative negotiation model based on system dynamics to quantitatively analyze risk allocation problems [6]. Karakas developed a multiagent system to simulate risk sharing and cost allocation among all parties [7]. Francesca constructed a game model of final quotation arbitration to estimate different risk sharing proportions of participants [8].

RISK IDENTIFICATION METHOD AND PROCESS UNDER BIM-BASED IPD PATTERN
As the primary link of risk management, risk identification refers to a comprehensive judgment, systematic classification, and scientific appraisal process of types and influence degrees of risk factors possibly existing in projects using IPD pattern. Project risk identification acquires and processes risk information mainly through two aspects: judge through professional knowledge and working experience of related professionals and then acquire information; obtain risk factors by summarizing related data and cases of actual projects. Common project risk identification methods include brain storming method, Delphi method, flowchart method, etc. Delphi method and flowchart method were mainly utilized to perform risk factor identification of the existing IPD projects according to actual requirements.
During the bidding phase of the IPD project, the design unit, construction unit and the owner reached an agreement on the delivery method, that is, an IPD team was formed to identify the main participants of the project. After the owner, design unit and construction unit signed a tripartite contract, the construction unit intervened in the design phase in advance to determine the project risk. The three parties will jointly discuss and determine the main subcontractors of mechanical and electrical, electric power, fire protection, and pipelines, and ensure that important subcontractors agree to add people to the project at the design stage, and jointly bear risks and responsibilities, and finally reach management goals and collect risks information. However, some subcontractors with relatively light responsibilities such as laying floors and carpentry still use traditional project management methods, sign total price contracts, and finally determine the estimated project risk situation. The process of forming an IPD team is shown in Fig. 1.

Figure 1
The IPD team formation process for construction projects [7] The risk identification process of IPD projects is usually divided into four phases: determination of main project participants, confirmation of project risks and management objectives, collection of risk information and estimation of project risk situation. However, the above steps can be combined in the actual implementation process, so the risk identification was carried out according to actual demands in this paper as shown in Tab. 1 and Tab. 2.  Geological and climatic condition a3 Turbulent political scene a4 Delay of project review and approval a5 Change of laws and regulations a6 Inflation a7 Tax change a8 Change of exchange rate a9 Project planning a10 Contract risk a11 Application of BIM technology a12 Design risk a13 Communication risk among parties a14 Too high early-stage expense a15 Engineering change a16 Construction quality a17 Construction schedule a18 Construction safety a19 Material supply a20 Insufficient experience of participant a21 Income is lower than the expected a22 Change of market demand

SCREENING OF RISK FACTORS OF COMPLEX PROJECTS UNDER BIM-BASED IPD PATTERN
These risk factors are not only existing risk factors, which are the same as those under the traditional project pattern, but also specific risk factors under IPD pattern. Finding out the primary risk factors influencing smooth completion of IPD projects is the fundamental work of risk management. The corrected Delphi method was utilized in this paper for scoring of risk factors in Tab. 2, and the scoring results are listed in Tab. 3 [9]. Risk factors of IPD projects were reduced and screened out based on the attribute reduction theory of rough sets, and the main risk factors influencing smooth implementation of IPD projects were obtained.

Basic Concept of Rough Set
Neighborhood rough set is established on the basis of classical rough sets. The classical rough set was proposed by Professor Z. Pawlak from Warsaw University of Technology in Poland when studying expression and utilization of incomplete data and inaccurate knowledge in 1982 [10]. Belonging to a basic theory in the data mining field, it can objectively mine internal information of data and extract data according to certain principles so as to obtain their connotation information.
Definition 1 [10]: Any subset of (relation) A × B is a binary relation from set A to set B, and it is marked as R, Definition 2: The non-null finite set is constituted by discussion objects we are interested in a domain of discourse and marked as U. If domain of discourse U and a cluster of equivalent relation R are given, they will form a tuple [10].
Definition 3 (indistinguishable relation): A knowledge base K = (U, R) is given, if P ⊆ R and P ≠ ∅, ∩ P is still an equivalent relation on domain of discourse U, and it is called indistinguishable relation on P and marked as IND(P), which is simplified as P [10].
Definition 4 (upper approximation and lower approximation of set): If a knowledge base is given, the domain of discourse U can be of multiple divisions (marked as U / IND(P)) and simplified as (U / P) according to indistinguishable relation determined by the subsets [10].
( ) R X is called lower approximation of X and ( ) R X is upper approximation of X. The boundary domain of subset X can be obtained as below: Definition 5 (rough set and accurate set): , set X is called an accurate set about domain of discourse -U relative to knowledge (equivalent relation) R.
, set X is a rough set about domain of discourse U relative to knowledge (equivalent relation) R [10]. For a decision system DS = (U, C ∪ D, V, f), its internal functions are respectively: U is domain of discourse, C is condition attribute, D is decision attribute, and C ∩ D = ∅, D ≠ ∅, and V is set of attribute values V a .
→ is an information function, which represents mapping relation among samples, attributes and attribute values.
A decision system DS is given, and the dependence degree of B C ∀ ⊆ decision attribute D on condition attribute subset B is defined as below:  Definition 7 (kernel of knowledge): Given a knowledge base K = (U, R) and a cluster of equivalents on it, P ⊆ R for any Q ∈ P, if Q satisfies [10].
It is said that Q is necessary in P, the set constituted by all necessary knowledge in P is the kernel of P, denoted as Core(P). Definition 8 (relative reduction of decision system): Given a decision system DS = (U, C ∪ D, V, f) [10]. The condition attribute subset B is called a relative reduction of condition attribute set C. Definition 9 (metric): In a given N-dimensional real number space is Ω, ∆ = R N × R N → R, then ∆ is called a measure (distance) on R N , if ∆ meets the following conditions: ∆(x 1 , x 2 ) ≥ 0, where and only when x 1 = x 2 , it is equal, function that represents the distance between the element x i and the element x j . The common distance function is as follows: Manhattan distance function: Euclidean distance function: P norm distance function: It can be concluded that the approximate boundary of X is as follows: Definition 12: Given a neighborhood decision system ( , ) NDS U A D = ∪ , the decision attribute D divides the domain of discourse U into N equivalence classes 1 ( , ..., ) n X X B A ∀ ⊆ the upper and lower approximations of the decision attribute D on subsets B are as follows: Similarly, the boundary of the decision system can be obtained as:   3 2 In the data processing of neighborhood rough sets, data will present differences in order of magnitudes and dimensions by the definition of rough set [11]. In order to obtain accurate data processing results, original data should be normalized before data processing of Eq. (14). min max min ( ) ( 1, 2, ..., ) i i For definition 12, the standard deviation of normalized attributes a 1 , a 2 , …, a 22 is solved as below: λ is a set parameter used to adjust neighborhood size according to data classification accuracy.

Reduction Results
Data in Tab. 3 are reduced to obtain 183 groups of reduction results. Different reduction sets are selected in projects under IPD pattern according to nature of participating enterprises and the definition of rough set, a i means the risk factors, a 5 (risk of laws and regulations), a 10 (contract risk), a 13 (communication risk between participants), a 15 (engineering change risk), a 20 (risk of insufficient experience of participants) and a 22 (change risk of market demands) are core elements of risks for construction projects under IPD pattern. Y -axis means the weight of a i. as is shown in Fig. 2.

Figure 2
The core elements of risks for construction projects under IPD pattern [11] The owner proposed new requirements when the steel structure of the health center was designed, hoping to change the back-to-back ward into the same direction. At the end of the design phase, when the owner proposes a design change temporarily, considering that the main subcontractor may have purchased equipment and allocated labor, the design unit will undergo a series of processes such as feasibility analysis after the modification is made. It is difficult for traditional management methods to quickly change ideas and make new adjustments. With the inherent flexibility of the IPD team, the design process of the design unit and the construction unit is a unified process, and the main contractor knows the progress of the project. For the latest ideas proposed by the owner, after the entire IPD team believes that it is feasible, the solution can be adjusted in a timely manner without increasing cost or time, so that the hospital's facilities can better meet customer requirements.

CONCLUSION
Risk factor sets, which possibly appeared in IPD projects, were evaluated and calculated through the attribute reduction algorithm based on the rough set theory. Core elements of risk factor sets were obtained as risk of laws and regulations, contract risk. Communication risk among participants, engineering change risk, risk of insufficient experience of participants and change risk of market demands, weights of risk factors were calculated through attribute frequency as shown in Tab. 3 and Tab. 2, and the reasonability of attribute reduction results was verified. The contributions of this article are as follows. Based on the improved neighborhood rough set theory, the risk factors of complex projects under the BIM-based IPD model are reduced and selected, and the most important risk factors for the successful completion of complex projects under the BIM-based IPD model are obtained. No discretization is required for continuous data, the original data attributes can be maintained, and the accuracy of rough set theory analysis can be improved. As there are relatively few laws and regulations and models of contract related to IPD pattern at present, risk of laws and regulations and contract risk have become the main problems faced in the complete and smooth implementation of IPD projects.