A sensor-cloud system is a combination of wireless sensor networks and cloud computing that is equipped with ubiquitous physical sensing ability, high-speed computation, huge storage, and so on. However, sensor-cloud systems suffer from various types of malicious attacks that can cause sensor communications to become unreliable. Establishing a trust evaluation method to ensure members' reliability in sensor cloud is an effective way to resist malicious attacks. However, most current trust evaluation methods are constrained to specific attacks or applications, and they lack compatibility, verifiability, and scalability. To solve these problems, we formulated the trust evaluation issue as a multiple linear regression problem. Considering energy restrictions, we adopt fog nodes to assist in the trust computation. Moreover, the least squares algorithm is used to find the fitting function between the communication feature and the trust value. The experimental results show that our approach can find the best trust evaluation model and improve the compatibility, verifiability, and accuracy of trust evaluation.