Short-term scheduling rule extraction method based on variable structure deep learning framework

Abstract

The invention discloses a short-term scheduling rule extraction method based on a variable structure deep learning framework. The method comprises the following steps: step 1, constructing a deep learning network; 2, selecting an input factor decision variable sample pair; 3, normalizing the sample data in the step 2; step 4, selecting and optimizing key hyper-parameters; step 5, and carrying outnetwork reconstruction; and step 6, network training, wherein the reconstructed network obtained in the step 5 is utilized to input the sample obtained in the step 3 for network learning, a final deeplearning network is obtained, and the network is the extracted short-term scheduling rule. Based on actual historical operation data of a power station, a deep learning network model based on a long-short-term memory network is constructed, internal rules contained in an actual operation process are mined, a power station short-term scheduling rule is established, and initial and end water levelsof a power station scheduling period, a period water incoming process and a power station power receiving network load process are used as input factors. The end water level of the power station period serves as a decision variable, and the model output result is more suitable for the actual scheduling process.

Description

technical field [0001] The invention relates to the field of hydropower energy optimization, in particular to a short-term scheduling rule extraction method based on a variable structure deep learning framework. Background technique [0002] The deterministic optimal dispatching of hydropower stations takes the future water and power grid load as a definite time series input. However, both runoff forecasting and power grid load forecasting have the characteristics of limited accuracy and limited forecast period. Difficult to apply well. [0003] The main idea of ​​implicit stochastic optimal scheduling is to extract the physical causal relationship between input factors and scheduling decisions by mining the inherent laws contained in the long sequence results of deterministic optimal scheduling, omitting the complicated model solving process, and then forming scheduling rules. However, since the training data source of this method is still the optimal scheduling result, it...

AI Technical Summary

Problems solved by technology

However, since the training data source of this method is still the optimal scheduling result, it is difficult to obtain the training sample data of the model, which also makes it difficult to apply the obtained scheduling results to the actual production process.

Images