Scholarlink

The prominence of new energy as a main source of social electricity supply is one of the primary manifestations of the clean and low-carbon characteristics of the new power system. Virtual Power Plants (VPPs) can effectively organize distributed new energy resources to participate in electricity market transactions as new types of entities and capture market dividends. Addressing the issue of low enthusiasm among VPP participants like distributed photovoltaics (PV) and energy storage systems (ESS) due to unstable returns, this paper proposes a joint forecasting and dynamic feedback optimization model for internal electricity quantity and price within a VPP. Firstly, considering the forecasting accuracy and physical rationality of internal electricity quantity and price, a multi-task deep neural network is constructed and improved, utilizing Bidirectional Long Short-Term Memory (BiLSTM) to simultaneously extract spatiotemporal features of quantity and price, and differentially modeling physical constraint embedding requirements such as PV multi-mode output and ESS cost decay. Secondly, a dynamic incentive feedback mechanism is designed and combined with online optimization based on game theory. Using operational cost as a benchmark, a closed-loop optimization framework for quantity-price forecasting is formed by integrating ma-chine learning, time series analysis, and physical modeling. To solve the Pareto equilibrium problem of participant economics, supply-demand deviation iteration combined with gradient descent is introduced to adjust forecast parameters, learning rate, and gradient in real-time, enabling real-time correction and regulation of electricity quantity and price forecasts. Finally, validation based on actual operational data from a provincial VPP shows that compared to traditional game theory models, the proposed model improves the forecasting accuracy of electricity quantity for PV and ESS participants by 1%-4% and price forecasting accuracy by 5%. This model accurately matches entity operational characteristics through coupled quantity-price forecasting, and resolves the “centralized optimization - individual loss” contradiction through dynamic price adjustment, providing sup-port for the sustainable increase of VPP entity revenue and participation enthusiasm under high penetration of new energy.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on a Joint Quantity-Price Forecasting Model Based on Virtual Power Plant Entity Revenue Optimization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xiaodong Cheng,
  • Baoshi Wang,
  • Jun Qiao,
  • Lanxu Wu,
  • Pengdong Tian

摘要

The prominence of new energy as a main source of social electricity supply is one of the primary manifestations of the clean and low-carbon characteristics of the new power system. Virtual Power Plants (VPPs) can effectively organize distributed new energy resources to participate in electricity market transactions as new types of entities and capture market dividends. Addressing the issue of low enthusiasm among VPP participants like distributed photovoltaics (PV) and energy storage systems (ESS) due to unstable returns, this paper proposes a joint forecasting and dynamic feedback optimization model for internal electricity quantity and price within a VPP. Firstly, considering the forecasting accuracy and physical rationality of internal electricity quantity and price, a multi-task deep neural network is constructed and improved, utilizing Bidirectional Long Short-Term Memory (BiLSTM) to simultaneously extract spatiotemporal features of quantity and price, and differentially modeling physical constraint embedding requirements such as PV multi-mode output and ESS cost decay. Secondly, a dynamic incentive feedback mechanism is designed and combined with online optimization based on game theory. Using operational cost as a benchmark, a closed-loop optimization framework for quantity-price forecasting is formed by integrating ma-chine learning, time series analysis, and physical modeling. To solve the Pareto equilibrium problem of participant economics, supply-demand deviation iteration combined with gradient descent is introduced to adjust forecast parameters, learning rate, and gradient in real-time, enabling real-time correction and regulation of electricity quantity and price forecasts. Finally, validation based on actual operational data from a provincial VPP shows that compared to traditional game theory models, the proposed model improves the forecasting accuracy of electricity quantity for PV and ESS participants by 1%-4% and price forecasting accuracy by 5%. This model accurately matches entity operational characteristics through coupled quantity-price forecasting, and resolves the “centralized optimization - individual loss” contradiction through dynamic price adjustment, providing sup-port for the sustainable increase of VPP entity revenue and participation enthusiasm under high penetration of new energy.