Scenarios sequence generation is the basis of scenario analysis and optimization problem, and its accuracy directly affects the effectiveness of related analysis and optimization calculation. Based on this, a new method of generating new energy output scenarios with load level constraints based on artificial intelligence is proposed, significantly improving the accuracy of the generated scenarios and overcoming the difficulties without effective methods faced by the dispatch center. The process of the new method is as follows: firstly, the historical data expansion strategy based on Kalman gain information fusion technology is adopted，realizing the effective data expansion of corresponding historical samples. Then, based on the self-organizing mapping network, the historical daily load curves are clustered to obtain multiple daily load type clusters, and each of the new energy output sequences on the same day as the daily load curves within each load cluster is corresponding classified into clusters, and the source load clusters with the specific load level constraint are achieved. Finally, a state transfer matrix based on Markov chain （MC） daily type transformation relationship is constructed, and then a sequence of daily types in the near future is generated by rolling sampling. And a generative adversarial network （GAN） already trained on each of source-load cluster data is applied to generate 96-point source-load scenarios sequences for the corresponding daily types. The numerical experiments verify the effectiveness and advancement of the proposed method.