CN-122026391-A - High-frequency time-of-use electricity price energy storage optimization scheduling method and system based on photovoltaic load bi-prediction and multi-peak valley identification

CN122026391ACN 122026391 ACN122026391 ACN 122026391ACN-122026391-A

Abstract

The invention discloses a high-frequency time-of-use electricity price and energy storage optimization scheduling method and system based on photovoltaic load double prediction and multimodal valley identification, and relates to the technical field of intelligent energy management and energy storage optimization scheduling, wherein the scheduling method comprises data acquisition and pretreatment, high-frequency time-of-use electricity price and multimodal valley dynamic identification, photovoltaic load double prediction fusion, energy storage charge and discharge multi-objective optimization scheduling and scheduling strategy real-time execution and adjustment; the dispatching system comprises a data acquisition and preprocessing module, a photovoltaic load double-prediction module, a multi-peak valley dynamic identification module, an energy storage optimization dispatching module, a real-time monitoring and execution module and a user interaction and decision support module, wherein the energy storage charging and discharging intelligent optimizing model facing the high-frequency electricity price environment is established by constructing a photovoltaic and load double-prediction fusion frame and combining a multi-peak valley dynamic identification algorithm of 15-minute-granularity electricity price data, so that the fine dispatching of the user side energy storage system is realized, and the economic benefit and the photovoltaic absorption rate are remarkably improved.

Inventors

HUANG XIAO
SHAN CHANGJUN
YAN JIAQUAN
Di Kesong
JI YINGLIANG

Assignees

杭州海兴电力科技股份有限公司
深圳和兴电力科技有限公司
广东和兴电力科技有限公司
海南海兴国际科技发展有限公司
宁波恒力达科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260416

Claims (10)

1. The high-frequency time-of-use electricity price energy storage optimization scheduling method based on photovoltaic load double prediction and multimodal valley identification is characterized by comprising the following steps of: S1, acquiring future electricity price prediction data, weather prediction data, historical photovoltaic power generation data, load electricity utilization data and energy storage system state data, and processing the data to obtain standardized data; s2, based on the standardized electricity price data, adopting a sliding window smoothing combined with local extremum detection algorithm to identify a plurality of peak time periods and valley time periods in the day, and calculating the identification confidence coefficient of each time period; S3, constructing a load demand prediction model and a photovoltaic power generation prediction model, taking a load prediction result as an exogenous variable of the photovoltaic prediction model, outputting a future load prediction sequence and a photovoltaic power prediction sequence in a cooperative manner, and calculating a net load curve; s4, solving a Pareto optimal charge-discharge strategy set by adopting an improved multi-objective optimization algorithm in combination with a multimodal valley recognition result, double-prediction fusion data and an energy storage system constraint condition with the aim of minimizing the cost of electricity charge of a user, maximizing the benefit of the energy storage system and maximizing the photovoltaic absorption rate; And S5, selecting a recommendation strategy from the Pareto optimal solution set according to user preferences, issuing the recommendation strategy to the energy storage system for execution, updating the prediction data and the scheduling plan every hour, and triggering emergency re-optimization when the actual operation deviation exceeds a threshold value.
2. The high-frequency time-of-use electricity price energy storage optimization scheduling method based on photovoltaic load double prediction and multimodal valley identification of claim 1, wherein S2 comprises: sliding window smoothing is carried out by adopting a moving average filter, and the window size is adaptively adjusted according to the volatility of electricity price; Determining a significance threshold based on the electricity price standard deviation, and detecting a local maximum value and a local minimum value of the smoothed sequence; expanding the detected extreme points into time periods, and filtering out the time periods with too short duration; Combining peak-valley time periods which are adjacent in time and have the same type, so that the continuity of the time periods is improved; And calculating the identification confidence of each peak-valley period, and comprehensively considering the price deviation degree and the duration factor.
3. The high-frequency time-of-use electricity price energy storage optimization scheduling method based on photovoltaic load double prediction and multimodal valley identification of claim 1, wherein S3 comprises: The load prediction sub-module adopts an online learning frame based on a gradient lifting tree to construct a multi-scale hysteresis characteristic, a rolling statistical characteristic and a time characteristic, and generates a load demand prediction sequence for 24 hours in the future through a recursion prediction strategy; The photovoltaic prediction sub-module adopts a neural network layered interpolation model or a time sequence fusion converter model based on deep learning, takes meteorological data and a load prediction result as exogenous variables, and outputs a theoretical photovoltaic power and actual photovoltaic power double sequence through multi-scale time sequence decomposition modeling; The double-prediction fusion mechanism inputs a load prediction result as an exogenous variable of photovoltaic prediction, firstly, performs load prediction to obtain a load power prediction sequence of 24 hours in the future, then combines the load prediction result with weather prediction data to construct an exogenous variable matrix of a photovoltaic prediction model, then performs photovoltaic prediction to obtain a photovoltaic power prediction sequence of 24 hours in the future, and finally subtracts a difference value of the photovoltaic power prediction value from the load power prediction sequence to obtain a net load; calculating a prediction error based on the current actual observed value every hour by using a real-time error correction mechanism, and adjusting a future predicted value by adopting an exponential decay correction strategy, wherein the correction weight decays exponentially along with the predicted time span; The module outputs a load prediction sequence, a photovoltaic power prediction sequence and a net load prediction sequence for 24 hours in the future and outputs confidence intervals of all prediction values at the same time, and when the confidence coefficient is lower than a threshold value, the system sends out early warning.
4. The high-frequency time-of-use electricity price energy storage optimization scheduling method based on photovoltaic load double prediction and multimodal valley identification of claim 1, wherein S4 comprises: establishing a multi-objective optimized mathematical model of energy storage charge-discharge scheduling, wherein decision variables are charging power of each period of 24 hours in the future And discharge power The objective function comprises three mutually conflicting optimization targets, namely, minimizing the cost of electricity charge of a user, maximizing the benefit of an energy storage system and maximizing the photovoltaic absorption rate; Executing an improved NSGA-II algorithm to generate a Pareto optimal solution set; Selecting a recommended solution from the Pareto optimal solution set for a user to decide; In the execution process, the optimization algorithm is restarted to generate a rolling scheduling plan according to the actual photovoltaic output, load demand and electricity price data update prediction per hour; and dynamically adjusting algorithm parameters and user preference weights according to the evaluation result to realize closed-loop optimization.
5. The photovoltaic load bi-prediction and multi-peak valley identification-based high-frequency time-of-use electricity price energy storage optimization scheduling method according to claim 4, wherein the expression of an objective function for minimizing the cost of electricity charge of a user is as follows: ; Wherein the method comprises the steps of For the power of the electric network, Purchasing power for a power grid in a period t, wherein Price (t) is the electricity Price of the period t; is the load power; is photovoltaic power; Expression of objective function for energy storage system profit maximization: ; expression of the objective function of photovoltaic absorption maximization: 。
6. The photovoltaic load bi-prediction and multi-peak valley identification-based high-frequency time-of-use electricity price energy storage optimization scheduling method according to claim 5, wherein the constraint conditions comprise: SOC upper and lower limit constraints Wherein, the SOC refers to the residual electric quantity of the energy storage system; Charge-discharge power constraint And ; SOC continuity constraints Wherein And In order to achieve the charge-discharge efficiency, In order to store energy in the form of a capacity, Is a time interval; Power balance constraint 。
7. The photovoltaic load bi-prediction and multi-peak valley identification-based high-frequency time-of-use electricity price energy storage optimization scheduling method according to claim 6, wherein the improved NSGA-II algorithm comprises: adaptive crossover mutation operator, crossover probability Probability of variation Dynamic adjustment with evolution algebra: ; ; Wherein g is the current algebra, For the maximum number of algebra, , , , ; The layering initialization strategy comprises the steps of dividing an initial population into three layers, namely 30% heuristic solution based on peak-valley electricity price, 30% greedy solution based on photovoltaic prediction and 40% random solution; A time sequence constraint repair mechanism, which is to design a special repair operator aiming at SOC continuity constraint, when an individual violates the SOC constraint, the charge and discharge power is adjusted backwards from the violating time, if If the discharge power is decreased or the charge power is increased before the time t Then the charging power is reduced or the discharging power is increased; dynamic penalty function method, for constraint violation, using dynamic penalty coefficients: ; Wherein, the As a result of the initial penalty factor, In order to penalize the growth exponent, For the degree of violation of the ith constraint, Is the total number of constraints; multi-objective weight adaptation, introducing user preference parameters (Satisfy the following ) And supporting interactive decision, wherein a user can adjust the weight of each target according to actual requirements, and the system calculates a weighted objective function in real time and recommends an optimal solution.
8. The high-frequency time-of-use electricity price energy storage optimization scheduling method based on photovoltaic load double prediction and multimodal valley recognition according to claim 7, wherein the algorithm flow of the improved NSGA-II algorithm is as follows: The method comprises the steps of firstly generating an initial population, then entering an iterative loop, wherein each generation carries out the following operations of evaluating an objective function value and constraint violation degree of each individual in the population, carrying out rapid non-dominant sorting to divide the individual into different levels, calculating the crowding degree distance of the individual in each level, generating a parent through tournament selection, carrying out self-adaptive crossover and mutation operation to generate a child, applying a time sequence constraint repair mechanism to ensure the feasibility of the child, combining the parent and the child to form a new population, selecting the next generation population according to the non-dominant level and the crowding degree distance, terminating after algorithm iteration maximum algebra or Pareto front convergence, and outputting a Pareto optimal solution set, wherein each solution corresponds to a complete 24-hour charge-discharge power sequence.
9. The photovoltaic load bi-prediction and multi-peak valley identification-based high-frequency time-of-use electricity price energy storage optimization scheduling method according to claim 8, wherein the recommendation solution comprises: Calculating normalized distance from each solution to ideal point, and selecting the solution with the smallest distance: ; Wherein, the And Respectively the minimum value and the maximum value of the jth target in the Pareto solution set; A single-target priority solution, namely selecting a solution with the optimal target according to a priority target designated by a user; weighted preference solution based on user-provided user preference parameters Calculating a weighted objective function and selecting an optimal solution: 。
10. a high-frequency time-of-use electricity price energy storage optimization scheduling system based on photovoltaic load bi-prediction and multi-peak valley identification, which is applicable to the scheduling method in any one of claims 1 to 9, and is characterized by comprising the following steps: The data acquisition and preprocessing module is used for acquiring real-time data and acquiring future electricity price prediction and weather prediction data from an external API; the photovoltaic load double-prediction module is used for generating future photovoltaic power prediction and load demand prediction according to historical data and weather prediction, and calculating a net load curve; The multi-peak valley dynamic identification module is used for analyzing the electricity price prediction data, identifying a plurality of peak valley time periods in the day and calculating the confidence coefficient; the energy storage optimization scheduling module is used for integrating the electricity price peak-valley information, photovoltaic load prediction and energy storage constraint, and running an improved NSGA-II algorithm to generate a Pareto optimal scheduling strategy set; the real-time monitoring and executing module is used for converting the optimal scheduling strategy into a specific control instruction and issuing the specific control instruction to the energy storage system for execution, and simultaneously monitoring the running state and performance index of the system in real time; and the user interaction and decision support module is used for selecting a recommendation scheme from the Pareto solution set according to user preferences.

Description

High-frequency time-of-use electricity price energy storage optimization scheduling method and system based on photovoltaic load bi-prediction and multi-peak valley identification Technical Field The invention relates to the technical field of intelligent energy management and energy storage optimization scheduling, in particular to a high-frequency time-of-use electricity price energy storage optimization scheduling method and system based on photovoltaic load double prediction and multimodal valley identification. Background With the acceleration of global energy transformation and the deep development of power market reform, a time-of-use electricity price mechanism has become an important means for guiding user-side demand response and promoting renewable energy consumption. High frequency time-of-use electricity price mechanisms with 15 minute granularity have been commonly implemented in developed areas such as europe, and electricity prices exhibit the complex characteristics of multimodal valleys and high fluctuation. Meanwhile, the rapid popularization of the distributed photovoltaic power generation system and the household energy storage equipment provides a new technical means for user side energy management. However, the existing energy storage scheduling technology is mainly aimed at the traditional peak-to-valley three-period electricity price mechanism, and the economic potential in the high-frequency electricity price environment is difficult to fully mine. In addition, the randomness of photovoltaic power generation and the volatility of user loads present significant challenges to the accurate scheduling of energy storage systems. The name "a dynamic time-of-use electricity price optimization decision method and system taking into account demand response" (publication number CN 116934374A) discloses an electricity price decision scheme. The scheme has the advantages that the time interval division can be adaptively adjusted according to the dynamic change of the load curve, and the scheme is more flexible than the traditional static peak-valley time interval division. However, the scheme mainly focuses on the establishment and optimization of electricity prices, focuses on the view angle of the power grid side or the electricity selling side, and is insufficient in research on the charge and discharge scheduling strategy of the user side energy storage system. In addition, the time division of the scheme is still based on a daily load curve, the multimodal valley characteristics under the environment of high-frequency electricity price (such as 15-minute granularity) are not considered, the fusion optimization of photovoltaic power generation and load prediction is not involved, and the scheme is difficult to adapt to a high-frequency time-of-use electricity price mechanism in areas such as Europe. The scheme is innovative in that the influence of various time-of-use electricity price schemes on energy storage scheduling is considered, and a reference can be provided for investment decision of an energy storage system. However, the main objective of this approach is to evaluate the benefit potential of energy storage systems under different electricity price strategies, belonging to post-hoc evaluation and analysis tools, rather than real-time scheduling optimization methods. The scheme does not relate to photovoltaic power generation prediction and load prediction, does not carry out deep research on a real-time scheduling strategy in a high-frequency electricity price environment, and cannot provide a refined charge and discharge control instruction for a user side energy storage system. The scheme is named as an optimization method of demand side time-of-use electricity price based on photovoltaic grid-connected uncertainty (publication number CN 106532769B), and has the advantages that the influence of the photovoltaic grid-connected uncertainty on electricity price optimization is considered, and the decision risk caused by the uncertainty is reduced through an opportunity constraint theory. However, the optimization object of the scheme is still the formulation of the time-of-use electricity price, and not the scheduling strategy of the user side energy storage system. Although uncertainty of photovoltaic power generation is considered in the scheme, a photovoltaic power prediction model and a load demand prediction model are not established, and are not subjected to deep fusion for energy storage scheduling optimization. In addition, the time division of the scheme is still based on the traditional peak-to-valley time period concept, and the research on the dynamic identification of the multi-peak valley in the high-frequency time-of-use electricity price environment is not conducted. The technical scheme has a certain progress in time-sharing electricity price optimization and energy storage scheduling, but most of the common problems are that an optimization object is mainly conc