Building photovoltaic data interpolation method based on WGAN and whale optimization algorithm

A data completion method for improving operational efficiency and reliability of a building photovoltaic generation system, characterized by utilizing a hybrid model based on a Wasserstein Generative Adversarial Network (WGAN) and a Whale Optimization Algorithm (WOA) to achieve completion of missing historical roof photovoltaic output data, the method comprising:

• obtaining historical building roof photovoltaic output data with a sampling frequency of 30 minutes over a one-year duration, and subjecting the data to preprocessing including cutting a long time series to separate daily load time series;
• forming 1*T*n tensor by using the separated daily load time series, where T represents a time step and n represents total number of features;
• and normalizing the daily load time series;
• constructing a Generative Adversarial Network (GAN) using a Convolutional Neural Network (CNN), wherein a generator G and a discriminator D learn a true distribution of the data through adversarial training, and utilizing a binary mask matrix to describe missing value locations of the data to guide the WGAN in generating missing data;
• optimizing input noise of the GAN generator by a Whale Optimization Algorithm (WOA), wherein Wasserstein distance is set to define a loss function of the generator G and the discriminator D, the WOA utilizes a total loss function comprising components from the generator G, discriminator D, and reconstruction losses as a fitness function, and performs the WOA to obtain optimized input noise capable of enabling the GAN to generate realistic and consistent reconstructed samples:
• the WOA comprises encircling prey, bubble net attack and searching prey,
• wherein calculation of encircling prey comprising:
• D
• →
• =
• ❘
• "\[LeftBracketingBar]"
• C
• →
• ⁢
• X
• →
• *
• (
• t
• )
• -
• X
• →
• (
• t
• )
• ❘
• "\[RightBracketingBar]"
• X
• →
• (
• t
• +
• 1
• )
• =
• X
• →
• *
• (
• t
• )
• -
• A
• →
• ·
• D
• →
• where t denotes total number of iterations, {right arrow over (A)} and {right arrow over (D)} denote coefficient vectors, {right arrow over (C)} denotes random vector, denotes position vector of optimal solution at current position, and {right arrow over (X)} denotes current position of prey,
• wherein calculation of bubble net attack comprising:
• {right arrow over (X)}(t+1)={right arrow over (D)}t·ebl·cos(2πl)+{right arrow over (X)}*(t)
• wherein {right arrow over (D)}t denotes distance between whale and the prey, b denotes a spiral constant, and l denotes a random number in the interval between 0 and 1,
• wherein calculation of searching prey comprises:
• {right arrow over (D)}=|{right arrow over (C)}·{right arrow over (X)}rand−{right arrow over (X)}|
• {right arrow over (X)}(t+1)={right arrow over (X)}rand−{right arrow over (A)}·{right arrow over (D)}
• where {right arrow over (X)}rand denotes position vector of a head whale in whale population;
• inputting optimal input noise determined by the WOA to the generator G to generate highly realistic reconstructed photovoltaic output data;
• and
• applying the reconstructed photovoltaic output data to an operational management of the building photovoltaic power generation system.