The paper proposes a data-driven fault-tolerant control (FTC) strategy to construct and accommodate the bias on ambient temperature measurements in supermarket refrigeration systems. The bias, which is caused by direct or indirect exposure of the sensor to the sun, can have a significant impact on the refrigeration system's energy consumption. Based on analysis of the real data a comprehensive model of the bias is developed and then used to generate realistic scenarios for testing the proposed FTC method. The FTC method uses a feed forward Neural Network (NN) as a black box model. The model is trained by active injection of perturbation signals during the night operations. During the Monte-Carlo tests, the strategy was implemented in a Plug & Play manner, demonstrating that substantial energy savings can be achieved during summer periods.