Changes in ambient temperature affect the State of Charge(SOC) of lithium-ion batteries, requiring a temperature-robust SOC estimation method. Current model-based solutions often lack interpretability, and learning strategy-based solutions necessitate complicated frameworks and more computational resources. To achieve interpretability in an efficient approach while improving accuracy, this paper proposes SocNet, a neural network for SOC estimation that integrates domain knowledge. The internal structure of the neural network is adjusted based on domain knowledge, adding a cross-layer highway for important variables and introducing a lookback mechanism for historical prediction. After training at 0 °C, 25 °C, 30 °C, and 50 °C, SocNet was tested at 10 °C and 40 °C, achieving a mean absolute error of 0.42%.