Power multi-modal feature level fusion method considering data difference

The invention discloses an electric power multi-modal feature level fusion method considering data difference, and the method comprises the steps: taking an electric power multi-parameter time sequence and image data as fusion objects, and carrying out the feature extraction of the electric power multi-parameter time sequence and image data; then, using cosine loss to construct a feature assimilation objective function, and carrying out unified description on different features to solve the semantic difference problem; establishing a mapping relation between specific data and the sensing capability of the electric power target based on the cross loss entropy, and proposing a fusion weight decision-making mechanism on the basis to solve the problem of sensing capability difference; and finally, weighting different features and carrying out electric power target perception. According to the method, the problems of insufficient consideration of semantic difference and perception capability difference among multi