Low-Carbon Regulation of Rural Photovoltaic Energy Storage Systems Based on Cooperative Optimization of Hybrid Neural Networks

At present, with large-scale construction of distributed photovoltaic (PV) systems underway in Western China, rural power supply stations face challenges such as low local PV consumption rates, insignificant carbon reduction effects, and insufficient grid capacity—issues that complicate the optimization and regulation of energy storage equipment. To address these challenges, this study focuses on the actual needs of a rural power supply station serving an agricultural park in Western China and proposes a data-driven control framework for PV-storage systems. The framework involves three key steps: first, developing a CNN-LSTM-Attention hybrid neural network for time-series load forecasting, which effectively captures the non-typical wide-sense stationary characteristics of rural loads and achieves superior accuracy compared to standalone LSTM models; second, establishing a dual-objective optimization model with minimal PV curtailment as the primary goal and operational cost reduction as the secondary goal, solved using a Particle Swarm Optimization (PSO) algorithm with 500 particles and 100 iterations; and finally, deploying the control strategy via Node-RED for real-time data acquisition and equipment regulation. Field tests demonstrate that this approach increases daily PV self-consumption by 30% while reducing operational costs, successfully achieving low-carbon optimal scheduling for the target area and validating its effectiveness in low-carbon operation.