Optical Current Sensor (OCS) is very sensitive to temperature changes, and temperature changes lead to errors in its measurement, which makes it difficult to meet the requirements of power system metering. Therefore, accurate prediction of OCS measurement errors caused by temperature changes is of great significance for monitoring its operational stability and ensuring the safe operation of the power system. Since the OCS output current is strongly nonlinear due to the influence of temperature, this paper proposes a radial basis Koopman-Kalman prediction algorithm for nonlinear power systems, which solves the problem that the OCS output current is difficult to predict under the influence of temperature due to strong nonlinearity. Firstly, the nonlinear OCS output current state quantities are mapped into the high-dimensional space to form an extended state by the Radial Basis Function (RBF), and the extended state is decomposed by the Extended Dynamic Mode Decomposition (EDMD) algorithm to calculate the approximate Koopman-Kalman algorithm in the high-dimensional space. Koopman operator approximation matrix. Secondly, the approximated Koopman operator is used for batch prediction in the high-dimensional linear space. Finally, Kalman filtering is used to update the correction to the last prediction of the batch prediction to follow the state change of the system. The OCS temperature-current data obtained from experimental measurements are used for experiments, and the results show that the mean square error MSE of the prediction algorithm proposed in this paper is reduced by more than 90% in comparison with both the standard Koopman prediction and the LSTM prediction for different temperature variations, which proves the effectiveness of the proposed algorithm.