Renewable energy generation and power load forecasting are important in both advanced smart grid and sustainable development. Many RNN related methods were used in prediction of power generation time series data but they often fails to capture very long-term correlations in practice due to the vanishing gradient problem. We introduced a modified version of the LSTNet model, which incorporates CNN, LSTM, SKIP-LSTM and Dense components. This model captures both short-term patterns and addresses the issue of vanishing gradients when capturing long-term patterns. We applied this model to predict hydro power and grid load, and by comparing the MAPE, MSE, and MAE metrics, it is evident that the performance is superior to the commonly used LSTM, GRU, and SimpleRNN models in time series analysis of smart grid. The integrated model can be applied to the intelligence of renewable energy grids.