CN-122001022-A - Multi-energy system optimal scheduling method based on wind-solar uncertainty set

CN122001022ACN 122001022 ACN122001022 ACN 122001022ACN-122001022-A

Abstract

The invention belongs to the technical field of comprehensive energy system optimization scheduling, and particularly relates to a multi-energy system optimization scheduling method based on a wind-solar uncertainty set, which comprises data acquisition and preprocessing, dynamic construction of the uncertainty set, energy storage life decay modeling, DRL agent training and scheduling strategy generation and execution; the scheme provides a dynamic uncertainty set construction method integrating deep reinforcement learning, which is used for adaptively adjusting the central position, the shape structure and the confidence level of a Wasserstein ellipsoid set through real-time sensing of weather conditions, seasonal variation and evolution of a prediction error mode by a DRL (data center language) intelligent agent, improving new energy consumption capability and system operation economy, constructing a refined life loss model integrating cyclic aging and calendar aging, converting the model into a reward function with economic cost embedded into the DRL, and realizing collaborative optimization of short-term economy and long-term health of the system.

Inventors

ZHAO HONGYE
LIU XIAOMIN
ZHANG SHENGWEI
LIU TINGXI
LI JIE
He Qinsi
ZHAO SHENGNAN

Assignees

内蒙古农业大学

Dates

Publication Date: 20260508
Application Date: 20260204

Claims (3)

1. The multi-energy system optimal scheduling method based on the wind-light uncertainty set is characterized by comprising the following steps of: Step S1, data acquisition and preprocessing, namely collecting historical wind-solar prediction errors, load data and related data of equipment operation states, and preprocessing to obtain a historical data set; Step S2, dynamically constructing an uncertainty set, and using a DRL intelligent agent to learn the dynamic characteristics of wind-light prediction errors in real time to construct a self-adaptive Wasserstein uncertainty set; Step S3, energy storage life attenuation modeling, namely constructing a life loss model based on the discharge depth and cycle life attenuation of an energy system, and embedding the life loss model into the DRL intelligent body through a piecewise linearization technology; Step S4, training the DRL agent by simulating an operation sequence, and optimizing an energy storage charging and discharging strategy; And S5, generating and executing a scheduling strategy, applying the strategy trained by the DRL agent to actual scheduling, dynamically generating a scheduling instruction according to the real-time wind-light output and the system state, performing rolling optimization, and continuously updating the uncertainty set and the scheduling strategy.
2. The multi-energy system optimal scheduling method based on the wind-light uncertainty set, which is disclosed by claim 1, is characterized in that in the step S2, the uncertainty set is dynamically constructed, and the method specifically comprises the following steps: Step S21, in an initial set construction stage, experience distribution is constructed based on a historical data set, and an initial uncertainty set is defined based on Wasserstein ellipsoids, wherein the following formula is used: ; in the formula, Representing the wind-solar prediction error vector, For the number of wind farms, For the number of photovoltaic power plants, Represents a real number and is used to represent a real number, Representing the mean value of the error vector, As a covariance matrix of the error vector, An inverse matrix representing the covariance matrix, The transpose is represented by the number, Representing an initial set of uncertainties; s22, constructing a DRL state space, constructing a DRL agent based on a deep reinforcement learning model, learning a dynamic adjustment strategy of an uncertainty set, and constructing a state vector for the DRL agent based on a historical data set; step S23, DRL motion space design, defining motion vector of DRL Motion vector An adjustment parameter comprising an uncertainty set center, an uncertainty set shape, and a confidence level; step S24, designing a reward function, constructing a multi-objective reward function, wherein the following formula is adopted: ; in the formula, 、、 And As the weight coefficient of the light-emitting diode, In order to indicate the function, The time of day is indicated as such, Representation of The wind-solar prediction error vector at the moment, Indicating time of day The set of uncertainties to be constructed, Representing a set of uncertainties Is defined by the volume of (a), Is shown at the moment The adjustment parameters of the uncertainty set center, Is shown at the moment The uncertainty sets the adjustment parameters of the shape matrix, Is the euclidean norm square, Representing the square of the Frobenius norm, Is a reward function; Step S25, dynamically updating on line, acquiring state information of an energy system in real time, constructing a real-time state vector, and dynamically adjusting parameters of motion vectors by a DRL intelligent agent according to the real-time state vector, wherein the parameters comprise uncertainty set center position adjustment, uncertainty set shape adjustment and confidence level adjustment; Step S26, self-adaptive constraint, introducing a self-adaptive constraint mechanism to enable the updated uncertainty set to meet physical constraints, wherein the physical constraints comprise boundary constraints, positive qualitative constraints and volume constraints.
3. The multi-energy system optimization scheduling method based on the wind-solar uncertainty set, according to claim 2, is characterized in that in step S3, energy storage life attenuation modeling is carried out, and the method specifically comprises the following steps: Step S31, the aging mechanism is identified and classified, the aging mechanism of the energy storage equipment is divided into two types of cyclic aging and calendar aging, and modeling is carried out respectively; S32, constructing a circulation aging quantization model, counting equivalent circulation times under different DoDs by adopting a rain flow counting method, carrying out life loss quantization by combining the equivalent full circulation number, and establishing a relation between the equivalent full circulation number and the residual capacity to obtain accumulated circulation capacity loss; Step S33, integrating a calendar aging model, quantifying the influence of temperature on calendar aging by adopting an Arrhenius equation, and obtaining aging capacity loss; Step S34, modeling the total life loss cost, converting cycle aging and calendar aging into economic cost, embedding an optimization objective function, and calculating to obtain the life loss cost, wherein the formula is as follows: ; in the formula, Represents the initial investment cost of the energy system, To the total equivalent full cycle number at the reference life, The number of equivalent full cycles resulting from cyclical aging, The number of equivalent cycles for the calendar aging conversion, The energy storage life loss cost in the current scheduling period; And step S35, optimizing and embedding, namely combining a cyclic aging quantization model, a calendar aging model and total life loss cost to construct a life loss model, integrating the life loss model into a DRL intelligent agent, expanding a DRL state space, updating a DRL rewarding function by using a life loss penalty term, and applying an SOC safety boundary and a DoD limit after the action of the DRL is output.

Description

Multi-energy system optimal scheduling method based on wind-solar uncertainty set Technical Field The invention belongs to the technical field of comprehensive energy system optimal scheduling, and particularly relates to a multi-energy system optimal scheduling method based on a wind-light uncertainty set, which is suitable for regional or park-level comprehensive energy systems containing wind power, photovoltaics, thermal power, energy storage, electricity-to-gas (P2G) and demand response resources. Background Along with the deep advancement of the 'double-carbon' target, the permeability of renewable energy sources such as wind energy, solar energy and the like continuously rises, but prediction errors caused by strong randomness, intermittence and volatility are difficult to avoid, and the safe and economic operation of a multi-energy system is severely restricted. The traditional uncertainty collection method has three problems of static conservation, poor adaptability and difficult engineering landing, namely, a box type collection ignores space-time correlation to cause standby redundancy, a fixed ellipsoid or Wasserstein collection is difficult to respond to weather mutation and relies on static history distribution to track dynamic evolution of wind-solar prediction errors in real time, and a traditional scheduling model ignores an energy storage life attenuation mechanism and does not convert physical aging into economic cost, so that excessive charge and discharge are caused, equipment life is shortened, and long-term asset value is damaged due to one-sided pursuit of short-term benefits. Therefore, there is a need for a data-driven, compact, computationally efficient method for constructing uncertainty sets that characterize wind-solar correlations, and on that basis, design an engineering-floor-based multi-energy system optimization scheduling strategy. Disclosure of Invention Aiming at the problems that the traditional uncertainty set method has three problems of static conservation, poor adaptability and difficult engineering landing, the invention provides a dynamic uncertainty set construction method of fusion Deep Reinforcement Learning (DRL), wherein the DRL agent perceives weather conditions, seasonal changes and evolution of a prediction error mode in real time, the central position, the shape structure and the confidence level of a Wasserstein ellipsoid set are adaptively adjusted, the high coverage rate is ensured, the conservation of the set is obviously reduced, and therefore, the new energy absorption capacity and the system operation economy are improved, and the problems that the traditional scheduling model is neglected to convert physical aging into economic cost, leads to overcharging discharge, shortens equipment life and damages long-term asset value are solved. The invention provides a multi-energy system optimal scheduling method based on a wind-light uncertainty set, which comprises the following steps: Step S1, data acquisition and preprocessing, namely collecting historical wind-solar prediction errors, load data and related data of equipment operation states, and preprocessing to obtain a historical data set; Step S2, dynamically constructing an uncertainty set, and using a DRL intelligent agent to learn the dynamic characteristics of wind-light prediction errors in real time to construct a self-adaptive Wasserstein uncertainty set; Step S3, energy storage life attenuation modeling, namely constructing a life loss model based on the discharge depth and cycle life attenuation of an energy system, and embedding the life loss model into the DRL intelligent body through a piecewise linearization technology; Step S4, training the DRL intelligent agent through simulation operation, optimizing an energy storage charging and discharging strategy, and balancing the toughness and the energy storage service life of the system; And S5, generating and executing a scheduling strategy, applying the strategy trained by the DRL agent to actual scheduling, dynamically generating a scheduling instruction according to the real-time wind-light output and the state of the multi-source system, performing rolling optimization, and continuously updating an uncertainty set and the scheduling strategy. Further, in step S2, the uncertainty set is dynamically constructed, specifically including the following steps: Step S21, in an initial set construction stage, experience distribution is constructed based on a historical data set, and an initial uncertainty set is defined based on Wasserstein ellipsoids, wherein the following formula is used: ; in the formula, Representing the wind-solar prediction error vector,For the number of wind farms,For the number of photovoltaic power plants,Represents a real number and is used to represent a real number,Representing the mean value of the error vector,As a covariance matrix of the error vector,An inverse matrix representing the covariance matrix,The tran