Flexible load resources can respond to power grid dispatching quickly without significant impact on user comfort because of their rapid response and flexible regulation. As the core part of flexible load

air conditioning load can reduce the peak power demand through scientific control strategy

and then relieve the pressure of power supply. In view of the nonlinear and fuzzy characteristics of air conditioning load data

a model of air conditioning load prediction based on modal decomposition and neural network is proposed. First

Pearson correlation coefficients are used to construct similar weekly load sequences. Then the load is decomposed by adaptive noise complete set empirical mode decomposition and variational mode decomposition（VMD）. In the VMD section

the original time series signal is input into the VMD layer and decomposed into multiple eigenmode functions（IMFs）by the VMD algorithm. These IMFs are input into convolutional neural network respectively

and their local features are extracted by convolutional

activation and pooling operations. These feature vectors are then fed into a bidirectional long short-term memory network

which uses its bidirectional propagation capability to capture long-term dependencies in the sequence. The improved whale algorithm is used to optimize the hyperparameters

and the regulation potential of the load is further discussed on the basis of the output forecast load sequence. The experimental results show that this method not only has high forecasting speed and accuracy

but also can reveal the adjustment potential of load more clearly.