The processing of time series data is the key technical field of financial data analysis. With the continuous development of computing science, deep learning has a revolutionary impact on the traditional computing model. Among them, the generative adversarial nets GANs has achieved desirable results in the field of data generation. Revolving around the conditional generative adversarial nets cGANs, an effective Bi-LSTM generator, CNN discriminator and data processing method are designed in this paper. Also, the experiments on two economic datasets including the stock and commodity price are implemented. The results show that compared with the traditional model, the prediction performance of this research method witnesses a great improvement and it can be employed to better deal with the analysis task of non-stationary data, which is a significant point contributed by the research. In addition, the details associated with the generator mode and GANs model optimization are reported and discussed in combination with the actual situation of the experiment, and the existing problems are further explained and discussed.

时间序列数据的处理是财务数据分析的关键技术领域。随着计算科学的不断发展，深度学习对传统计算模型产生了革命性的影响。其中，生成对抗网络GAN在数据生成领域已取得理想的结果。围绕条件生成对抗网络cGAN，设计了一种有效的Bi-LSTM生成器，CNN鉴别器和数据处理方法。此外，还对包括股票和商品价格在内的两个经济数据集进行了实验。结果表明，与传统模型相比，该研究方法的预测性能有了较大的提高，可以更好地处理非平稳数据的分析任务，这是该研究做出的重要贡献。此外，结合发电机的实际情况和实际情况，对发电机模式和GANs模型优化相关的细节进行了报道和讨论，并对存在的问题进行了进一步的解释和讨论。