As an important part of the energy storage power station of novel power system, the lithium battery needs to estimate the battery power, that is, the state of charge in real time during its operation, so as to avoid the overcharge and over discharge failures of the energy storage lithium battery, which brings security risks to the energy storage system. In order to accurately estimate the battery state of charge (SOC), this paper tests and studies lithium batteries with different SOCs in a wide temperature range based on electrochemical impedance spectroscopy (EIS), and proposes a SOC evaluation method of lithium battery under wide temperature range based on EIS. The experimental results show that in the normal operating temperature range (15~45 ℃), the impedance amplitude in the 10~1000 Hz frequency band can effectively characterize the battery SOC at different temperatures, while the impedance phase basically does not change with the change of the battery SOC. By comparing the impedance amplitude and phase at different temperatures and different SOCs, this paper selects 10 Hz, 100 Hz, and 1000 Hz as the characteristic frequencies, and uses the impedance amplitude and ambient temperature at the three frequencies as input parameters. Combined with the deep neural network algorithm (DNN), the SOC estimation of lithium battery is realized. The results show that the SOC estimation model established in this paper can control the estimation error within 4%, which can effectively provide a reference for lithium battery SOC estimation and improve the equipment management level of the energy storage power station in novel power system.