A method for predicting regional energy load evolution based on deep reinforcement learning

By employing wavelet transform, causal dilated convolution, and graph topology techniques, combined with physical evolution equations and a two-layer asynchronous architecture, the problems of load prediction accuracy decay and scheduling strategy failure in multi-physics coupling scenarios in agricultural microgrids were solved, achieving efficient multi-energy system collaborative scheduling.

CN122175107BActive Publication Date: 2026-07-21SOUTHEAST UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SOUTHEAST UNIV
Filing Date
2026-05-13
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing technologies in agricultural microgrids suffer from accuracy degradation of data-driven load forecasting models and failure of scheduling strategies due to strong nonlinear state abrupt changes and long-period dynamic hysteresis effects within the system. This is especially true in scenarios with strong coupling of multiple physics fields, where effective decoupling and accurate prediction are difficult to achieve.

Method used

Wavelet transform and causal dilatation convolution are used to decouple multi-source data in the frequency domain and align spatiotemporal causality. Combined with graph topology and physical evolution equations, a two-layer asynchronous architecture is constructed. The model is updated using a time-difference credit allocation mechanism, and the output electrothermal scheduling action is associated with instantaneous and delayed reward signals.

Benefits of technology

It improves the physical and logical consistency and prediction accuracy of multi-source data fusion, ensures that the model has good strategy convergence and robustness when facing complex nonlinear thermodynamic states, and realizes the balance between transient electrical load stability regulation and long-term thermal environment economic operation of microgrids.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122175107B_ABST
    Figure CN122175107B_ABST
Patent Text Reader

Abstract

The application discloses a kind of district energy load evolution prediction methods based on deep reinforcement learning.The application utilizes wavelet transform to decouple multi-source monitoring data in frequency domain, and combines pipe network flow rate to execute causal inflation convolution translation, realize the causal alignment of electric heat data;Secondly, the device graph topology is constructed, and the phase change state and transient thermal impedance are mapped to the edge weight to generate the situation awareness graph;Then, the state vector is extracted by using the physical evolution equation to execute nonlinear constraint mapping and combining high-frequency electric transient feature embedding mechanism;Subsequently, the asynchronous period electric energy and thermal energy scheduling action is output by double-layer asynchronous architecture, and the immediate and delayed reward signal is associated;Finally, the model is updated by using the time difference credit distribution mechanism based on heat transfer loss rate.The application effectively overcomes the prediction accuracy decay and strategy failure problem caused by strong nonlinear state mutation and long-period dynamic hysteresis.
Need to check novelty before this filing date? Find Prior Art