The accelerated process of economic globalization and the urgent need for global economic recovery after the COVID-19 epidemic have led to increasing interconnectedness among countries. At the same time, the risk exposure among economies is expanding, and the financial risks embedded in each country itself will spread further through different countries and economic activities, which may lead to systemic risks. Most studies are based on single-layer financial networks, whereas relatively few studies are based on multi-layer financial networks, and existing studies do not consider the risk-averse behavior of each agent in the risky state. The financial system in the new situation presents new characteristics of multiple sources of risk, multiple transmission channels, and risk overlap under different transmission channels, making it necessary to study systemic risk from the perspective of multi-layer financial networks with multiple agents and multiple links.Based on complex network theory, this paper portrays three different financial agents, namely, banks, firms and assets, and constructs a multi-layer financial network system model based on the multiple links of bank-firm credit, interbank lending and bank-asset portfolios. Then, the risk-averse behavior of firms actively seeking loan assistance from potential creditor banks and the two risk-averse behaviors of banks actively selling assets and contracting credit are further considered. The impact of different agents’ risk-averse behaviors on systemic risk is simulated. Furthermore, based on the risk characteristics of banks and firms, different strategies under their active risk-averse behaviors and the superposition of multiple risk-averse strategies are fully considered to further explore the optimal risk-averse strategies.First, the states of both banks and firms are divided into three: no-stress, stress and collapse exit states. After suffering from external shocks, firms under stress can seek loan assistance from other potential creditor banks. When a bank is under stress, it will sell some of its assets or contract credit. The unit value of the sold assets will depreciate and trigger a fire-sale effect, thus causing other banks holding common assets to suffer some asset losses; the loan acquisition rate of the repossessed firms will also decrease. The risk under the above behavior will be transmitted to other banks through the credit linkage between banks and firms and the common portfolio.Then, the bank-firm credit network is constructed based on banks’ asset sizes, firms’ borrowing sizes, and the average connection degree of banks and firms. Based on the total asset value and their average connection degree, a bank-asset portfolio network is constructed; based on banks’ lending assets and borrowing probability functions, an interbank lending network is constructed. Based on the above bank-firm credit network, interbank lending network and bank-asset portfolio network, a multi-layer financial network system model is constructed with banks as the center. By setting certain external shocks to trigger banks to fail and liquidate, the risk is propagated in the system through multiple links, and each financial agent suffering from the risk shock adopts different active risk-averse behaviors, thus triggering further evolution of the multi-layer financial network system until there are no new failures in the system.By comparing the cumulative default probability of the banking system under different agents’ risk-averse behaviors and strategies, the impact of agents’ risk-averse behaviors on the systemic risk of the multi-layer financial network and the optimal risk-averse strategies are explored. It is found that firms actively seeking credit support from potential creditor banks can effectively inhibit the propagation of risk in multi-layer financial networks and eventually effectively mitigate systemic risk, whereas banks actively selling external assets or contracting credit can exacerbate systemic risk in multi-layer financial network systems. When considering the different risk-averse strategies of each agent, the mitigation of systemic risk in the multi-layer financial network is found to be optimal when firms preferentially seek credit from banks with higher capital adequacy ratios or when they preferentially sell assets with higher cumulative depreciation rates. Moreover, the effect is most significant when these two risk-averse strategies are superimposed. Finally, the reliability of the results of this study was demonstrated using a robustness test.