Systems and methods for skyline prediction for cyber-physical photovoltaic array control

1. A method for classifying a set of video features, comprising:

accessing a video representative of a first set of video features;

modeling the video representative of the first set of video features as a linear dynamical system;

obtaining an observability matrix for the linear dynamical system;

extracting a set of tangent vectors using the observability matrix, wherein the set of tangent vectors is representative of the linear dynamical system;

developing a probability density function representative of the linear dynamical system based on the set of tangent vectors;

assigning a score to a set of video features by evaluating the probability density function for each set of video features;

obtaining a spatiotemporal representation of the set of video features in terms of Lyapunov exponents; and

classifying a set of video features using a combination of the spatiotemporal representation and the score, wherein the largest Lyapunov exponents are representative of transition phases between a set of classes.

2. The system of claim 1 , wherein the set of tangent vectors lie on a Grassmannian manifold.

3. The system of claim 1 , wherein the probability density function lies on a Grassmannian manifold.

4. The system of claim 1 , wherein the linear dynamical system is driven by zero mean white Gaussian noise.

5. A method for predicting a set of future video features, comprising:

accessing a video, wherein the video is representative of a set of video features;

developing a time series matrix using a set of pixel values from the video;

developing a phase space matrix from the time series matrix, wherein the phase space matrix is representative of the time series matrix in phase space, wherein phase space models all possible states of a system; and

calculating a set of trajectories in phase space, wherein the set of trajectories in phase space is calculated by performing kernel regression on a weighted average of neighboring points in phase space, wherein the neighboring points are selected using locality sensitive hashing;

wherein the set of trajectories in phase space is representative of a predicted set of pixel values,

wherein the predicted set of pixel values is representative of a set of future video features.

6. The method of claim 5 , wherein the time series matrix comprises pixel values for a plurality of pixel locations in Cartesian space, wherein the pixel values vary over time.

7. The system of claim 5 , wherein a subject of the video comprises a skyline, wherein the set of video features is representative of the skyline.

8. The method of claim 5 , wherein the phase space matrix is developed using embedding parameters intrinsic to the time series matrix, wherein the embedding parameters comprise embedding delay and embedding dimension.

9. The method of claim 5 , wherein a minimum embedding delay is obtained by finding a lowest mutual information between samples expressed in the time series matrix.