Due to the massive deployment of electrical network devices, Fiber-Wireless (FiWi) access network has to suffer from the challenge of high energy consumption. With the ever-increasing eager for green communication infrastructure, the issue of high energy consumption may become one of the major barriers for the advance of FiWi access network. Previous works proposed for green FiWi access network cover three aspects, including the optimized deployment of network devices such as optical network units (ONUs), the energy-efficient bandwidth allocation of ONUs with QoS guarantee, and the dynamic power states scheduling of ONUs (i.e., active/sleep) according to their traffic profile. However, these works did not take the energy-saving design of wireless subnetwork into account simultaneously. In fact, when some ONUs in the optical subnetwork are switched into sleep states, part of the radio interfaces originally forwarding traffic to the sleep ONUs would be idle or low-loaded. This provides a potential opportunity for energy-saving in wireless subnetwork by switching off the idle or low-loaded radios. This paper focuses on the design of green multi-radio FiWi access network by integrating the energy-savings of optical and wireless subnetworks. To support the energy-efficient design, the new power states are defined for ONUs and radios, respectively. Aiming at dynamic traffic profile, the heuristic algorithms are proposed for the energy saving of integrated wireless front-end and optical back-end of FiWi access network. First, the Energy-saving algorithm with ONU sleep mechanism (EAS) is proposed to dynamically schedule the power states of ONUs by judging their traffic profile with load thresholds. Then, the Energy-saving algorithm based on Radios Off (ERO) is proposed to reconfigure the topology of wireless subnetwork by controlling the power states of radios dynamically. Moreover, wireless rerouting is employed in both EAS and ERO to guarantee the QoS provisioning ability of network. Finally, a comprehensive energy-saving scheme called EE is proposed by combining the EAS and ERO algorithms strategically. Simulation results show that with the reasonable setting of parameters, the proposed EE scheme can save the energy of 33.14% to 64.35% and 8.56% to 36.42% in a wide range of traffic load compared to the No-energy-saving and QoS-aware energy-saving scenarios, respectively.