Renewable-energy-based grids development needs new methods to maintain the balance between the load and generation using the efficient energy storages models. Most of the available energy storages models do not take into account such important features as the nonlinear dependence of efficiency on lifetime and changes in capacity over time horizon, the distribution of load between several independent storages. In order to solve these problems the Volterra integral dynamical models are employed. Such models allow to determine the alternating power function for given/forecasted load and generation datasets. In order to efficiently solve this problem, the load forecasting models were proposed using deep learning and support vector regression models. Forecasting models use various features including average daily temperature, load values with time shift and moving averages. Effectiveness of the proposed energy balancing method using the state-of-the-art forecasting models is demonstrated on the real datasets of Germany's electric grid.