Individual load curves generation and personal data protection

With the deployment of smart meters, the nature and volume of consumption data managed by the French DSO Enedis increase significantly. The challenge for Enedis is to develop a "secure" solution that guarantees the confidentiality of its customers' consumption data while meeting the following objectives: actively contributing to the energy transition by providing meaningful data to states and local authorities; providing curves with enough information to be useful for the needs of the concerned public; and, complying with personal data protection laws, thus protecting our customers' privacy (GDPR). DeepCourboGen is a tool designed to simulate individual electricity load curves using "deep learning" techniques, conditioned by client characteristics. It allows for the generation of annual load curves with 30-minute time steps. The Deep Convolutional Wasserstein Generative Adversarial Networks (DC-WGAN) method trains the simulator, while Siamese networks detect similarities between real and generated curves. Multiple metrics measure performance like mean, coefficient of variance, skewness, kurtosis, Siamese correlation. Efficiency and performance of the DeepCourboGen are improved by fine-tuning techniques. Computational resources required (time and memory) are reduced while maintaining or even improving the accuracy of the generated load curves making the solution deployment more scalable and more practical for real-world applications.