The prediction and optimization of distribution network asset costs is a complex problem in the power industry, involving the optimization of multiple objectives and the response to dynamic demands. Traditional methods often struggle to effectively address fluctuations in power load and the uncertainties in the supply chain, limiting their effectiveness in complex environments. To solve this issue, we propose the TransGrid-CostOpt model, an intelligent cost optimization model that integrates deep learning, multi-objective optimization, time-series forecasting, and optimization decision-making modules. TransGrid-CostOpt optimizes load forecasting and cost allocation for the distribution network by combining multi-source data, time-series load forecasting, and reinforcement learning decision strategies, reducing operational costs, improving load forecasting accuracy, and enhancing decision adaptability. Experimental results show that TransGrid-CostOpt outperforms traditional models and other advanced methods on the BuildingsBench and PJM Hourly Load Data datasets, exhibiting higher accuracy and efficiency in forecasting, cost optimization, and multi-objective balancing. Compared to classical baseline models and cutting-edge approaches, TransGrid-CostOpt demonstrates a 15% to 30% overall performance improvement. Ablation experiments confirm the critical role of each module, especially the time-series forecasting module and optimization decision-making module, in significantly enhancing the model’s performance. TransGrid-CostOpt strengthens the cost management capability of the distribution network and shows strong adaptability in dynamic electricity market environments, with broad application potential.
Xiong et al. (Fri,) studied this question.