关键词: 项目组合网络, 风险级联传播, 韧性测度, 改进贝叶斯网络

Abstract: Due to shared resources, similar technology and process requirements, overlapping target markets, and the diffusion of knowledge or experience among projects, there exist dependencies among various projects within the project portfolio. Also, because of these dependencies, risks occurring in one project can be transferred to other projects, potentially leading to the failure of the entire project portfolio. Aiming at the multi-sourced risk in complex R&D projects and risk propagation between projects, the paper builds a model of risk propagation considering multi-sourced risk and multi-stated problems caused by risk propagation based on the improved Bayesian network model.
Firstly, the paper analyzes the multi-sourced risks and multi-stated problems in complex R&D project. Multi-sourced risks refer to different types of risks that exist during the process of project portfolio, such as technological risks, management risks, business risks, and external risks. Multi-stated refer to changes in project status caused by the occurrence of risks and the cascading propagation of risks within the project portfolio network. Based on the indicators of “the probability of risk occurrence” and “the probability of risk diffusion”, the paper classifies the projects in the portfolio into four states, specifically: 1)The risk transferrer, which refers to the projects with a high probability of risk occurrence and a high probability of risk diffusion, indicating that the project is prone to risks and is likely to transfer these risks to other projects in the network. 2)The risk terminator, which refers to projects with a high probability of risk occurrence but a low probability of risk diffusion, indicating that the project has encountered risks, but the team has excellent risk-handling capabilities and has effectively resolved these risks. And, the project accumulates rich knowledge and experience in the process and ensures that the risk does not transfer to other projects in the network. 3)The risk immunizer, which refers to projects with a low probability of risk occurrence and a low probability of risk diffusion, indicating that the project has not yet encountered risks. Additionally, the project team possesses excellent risk-handling capabilities, enabling the risks to be resolved within the project without transferring to other projects in the network. 4)The risk susceptible, which refers to projects with a low probability of risk occurrence but a high probability of risk diffusion, indicating that the project has not yet encountered risks, but the project team has poor risk-handling capacities, making it difficult to resolve risks within the project, thus increasing the likelihood of these risks transferring to other projects in the portfolio. In summary, the probability of risk occurrence depends on the project’s own risk probability and the probability of obtaining risks from other dependent projects due to risk propagation. And, the probability of risk diffusion depends on the risk-handling capabilities of project’s R&D team after the occurrence of risks.
Furthermore, the paper analyzes the risk propagation in the portfolio network through the construction of an improved Bayesian network. Bayesian network, a commonly used method in machine learning, is a process of determining posterior probabilities based on known conditional probabilities and prior probabilities. In traditional Bayesian network models, nodes only have two states: Failure(F) and True(T). However, for individual projects within the project portfolio, merely using failure and success to measure the project’s status is inaccurate. For projects within the project portfolio, it is necessary to measure the impact of “multi-sourced risks” and “multi-stated problems” on the risk propagation in portfolio. Therefore, the paper further constructs an improved Bayesian network model to analyze the risk propagation process of “multi-sourced risks” and “multi-stated”.
In addition, resilience is an important issue in the field of complex networks, which refers to the ability of the whole network to return to the initial state or better state when a certain element of the network is at risk. For the portfolio network, resilience refers to the ability of the entire portfolio to withstand the risk and achieve its initial performance when the project is exposed to risk. Under the condition of risk propagation, the risk influence includes direct impact and indirect impact. The direct impact is on the project where the risk arises, and the ability of the project to achieve its initial performance may be affected. Indirect impact refers to other projects in the portfolio. The risk may be transmitted to other projects in the network that are directly and indirectly related to the project, so that the performance of the whole project portfolio will be affected. In conclusion, the paper analyzes the resilience of project portfolio network through the following aspects: 1)The ability of the whole portfolio to achieve its initial performance after the occurrence of risks, that is, the robustness analysis of the project portfolio. 2)How quickly the portfolio can recover from the risk event, i.e. the time required for the portfolio to recover its initial performance. 3)The cost required to enable the entire project portfolio to achieve its initial performance after the occurrence of risks. Finally, the R&D project portfolio is taken as an example to verify the effectiveness of the model and method proposed in the paper.

Key words: the project portfolio network, risk propagation, resilience analysis, the improved Bayesian network