A light storage energy regulation method and system based on multi-time scale rolling optimization

Abstract

The application provides a kind of light storage energy regulation method and system based on multi-time scale rolling optimization in the technical field of energy storage regulation, and the method comprises the following steps: S1, input 24-hour electricity price prediction curve, 24-hour photovoltaic output prediction curve, 24-hour load prediction curve and initial SOC into day-ahead prediction model to obtain 24-hour SOC reference trajectory and benchmark charging and discharging plan; S2, collect feedback data, input 24-hour SOC reference trajectory, benchmark charging and discharging plan and feedback data into intra-day prediction model to obtain corrected charging and discharging plan; S3, set SOC dynamic limit, collect real-time electricity price and real-time photovoltaic output, input corrected charging and discharging plan, SOC dynamic limit, real-time electricity price, real-time photovoltaic output and actual SOC into real-time prediction model to obtain charging and discharging instruction and frequency modulation output power. The application has the advantages of greatly improving the accuracy, economy and grid adaptability of energy storage regulation, and greatly delaying the battery life attenuation speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage regulation, and particularly discloses a photovoltaic energy storage regulation method and system based on multi-time scale rolling optimization. BACKGROUND

[0002] With the vigorous advocacy and promotion of clean energy worldwide, the proportion of photovoltaic power generation in the power system is increasing. However, photovoltaic power generation is affected by natural conditions (such as light intensity, weather changes, etc.), and presents obvious intermittency and volatility, which brings serious challenges to the stable and reliable operation of the power system. As an effective means to solve this problem, the energy storage system can store electric energy when the photovoltaic output is excessive and release electric energy when the photovoltaic output is insufficient, thereby smoothing power fluctuations and improving the stability of the power system. At the same time, the real-time electricity price mechanism is widely used in the electricity market, so that the electricity price fluctuates in real time with the power supply and demand relationship. Therefore, there is a demand for the energy storage system to dynamically optimize the charging and discharging strategy according to the real-time electricity price signal and the photovoltaic output condition, in order to achieve efficient regulation. However, the current photovoltaic energy storage station faces the following core challenges:

[0003] 1. Time scale fragmentation problem: the traditional method uses a single time scale (such as only day-ahead optimization), which cannot take into account the medium and long-term electricity price trend and real-time fluctuation characteristics, resulting in a large deviation between the charging and discharging strategy of the energy storage system and the actual electricity price.

[0004] 2. Prediction error accumulation problem: the accuracy of photovoltaic output prediction and electricity price prediction is insufficient, and there is a lack of dynamic compensation mechanism, resulting in frequent adjustment of the charging and discharging of the energy storage system, accelerating the degradation of battery life.

[0005] 3. Economic and life contradiction problem: ignoring the cost of battery life degradation, excessively pursuing short-term benefits at the expense of the whole life cycle economy of the energy storage system.

[0006] 4. Insufficient grid coordination problem: the ability of the energy storage system to participate in frequency modulation, backup and other auxiliary services is not fully utilized, and the grid adaptability is poor.

[0007] Therefore, how to provide a photovoltaic energy storage regulation method and system based on multi-time scale rolling optimization to improve the accuracy, economy and grid adaptability of energy storage regulation, and to slow down the degradation rate of battery life, has become a technical problem to be solved. SUMMARY

[0008] The technical problem to be solved by the present application is to provide a photovoltaic energy storage regulation method and system based on multi-time scale rolling optimization to improve the accuracy, economy and grid adaptability of energy storage regulation, and to slow down the degradation rate of battery life.

[0009] In a first aspect, the present application provides a light storage energy regulation method based on multi-time scale rolling optimization, comprising the following steps:

[0010] Step S1, creating and training an electricity price prediction model, a photovoltaic output prediction model, a load prediction model, a day-ahead prediction model, an intra-day prediction model and a real-time prediction model;

[0011] Step S2, obtaining electricity price historical data, photovoltaic output historical data, load historical data and initial SOC of the energy storage system, inputting the electricity price historical data into the electricity price prediction model to obtain a 24-hour electricity price prediction curve, inputting the photovoltaic output historical data into the photovoltaic output prediction model to obtain a 24-hour photovoltaic output prediction curve, and inputting the load historical data into the load prediction model to obtain a 24-hour load prediction curve;

[0012] Step S3, inputting the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference trajectory and a benchmark charging and discharging plan;

[0013] Step S4, collecting feedback data including actual SOC, photovoltaic output deviation and frequency modulation response record, inputting the 24-hour SOC reference trajectory, the benchmark charging and discharging plan and the feedback data into the intra-day prediction model to obtain a corrected charging and discharging plan;

[0014] Step S5, setting a SOC dynamic limit value, collecting real-time electricity price and real-time photovoltaic output, inputting the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC into the real-time prediction model to obtain charging and discharging instructions and frequency modulation output power, and regulating the energy storage based on the charging and discharging instructions and the frequency modulation output power;

[0015] Step S6, rolling optimizing the output of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model;

[0016] Step S7, setting a deviation compensation mechanism, and compensating the frequency modulation output power based on the deviation compensation mechanism.

[0017] Further, in the step S1, the electricity price prediction model is constructed based on a first LSTM network, and is used to output a 24-hour electricity price prediction curve according to electricity price historical data;

[0018] The photovoltaic output prediction model is constructed based on an NWP meteorological network and an XGBoost network, and is used to output a 24-hour photovoltaic output prediction curve according to photovoltaic output historical data;

[0019] The load prediction model is constructed based on a second LSTM network, and is configured to output a 24-hour load prediction curve according to historical load data.

[0020] Further, in the step S1, the day-ahead prediction model is configured to output a 24-hour SOC reference trajectory and a benchmark charging and discharging plan based on the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and the initial SOC, with 1 hour as the resolution; and the 24-hour SOC reference trajectory is used as a boundary condition of the intra-day prediction model.

[0021] The objective function output by the day-ahead prediction model is to maximize the value of the whole-cycle profit function, and is constrained by the energy storage dynamic constraint function, the grid interaction constraint function, the charging and discharging mutual exclusion constraint function and the photovoltaic output uncertainty robust optimization constraint function.

[0022] The formula of the whole-cycle profit function is as follows:

[0023]

[0024] wherein, Max Profit DA represents the maximum value of the whole-cycle profit of the energy storage system; λ DA (t) represents the electricity price prediction value of the tth hour of the day-ahead; P dischar (t) represents the discharging power of the energy storage system in the tth hour of the day-ahead; P char (t) represents the charging power of the energy storage system in the tth hour of the day-ahead; Δt DA represents the time resolution of the day-ahead prediction model, i.e. 1 hour; C degradation represents the life loss cost based on the Rainflow counting method; C grid represents the grid interaction penalty term, which is configured to suppress the power peak-valley fluctuation;

[0025] The formula of the energy storage dynamic constraint function is as follows:

[0026]

[0027] wherein, SOC(t) represents the SOC of the energy storage system in the tth hour of the day-ahead; SOC(t-1) represents the SOC of the energy storage system in the (t-1)th hour of the day-ahead; η char represents the charging efficiency of the energy storage system; and represents the discharging efficiency of the energy storage system; E rated represents the rated capacity of the energy storage system;

[0028] The formula of the grid interaction constraint function is as follows:

[0029] P grid,min (t) ≤ P grid (t) = Pload (t) + P char (t) - P pv_pred (t) - P dischar (t) ≤

[0030] P grid,max (t);

[0031] wherein, P grid,min (t) represents the minimum power output of the power grid at the tth hour of the day-ahead; P grid (t) represents the power of the power grid at the tth hour of the day-ahead; P load (t) represents the load power at the tth hour of the day-ahead; P pv_pred (t) represents the predicted value of the photovoltaic output at the tth hour of the day-ahead; P grid,max (t) represents the maximum power output of the power grid at the tth hour of the day-ahead;

[0032] The formula of the charge-discharge mutual exclusion constraint function is:

[0033] P dischar (t)·P char (t) = 0;

[0034] The formula of the photovoltaic output uncertainty robust optimization constraint function is:

[0035] P pv_min (t) ≤ P pv_pred (t) ≤ P pv_max (t);

[0036] wherein, P pv_min (t) represents the minimum value of the predicted value of the photovoltaic output at the tth hour of the day-ahead; P pv_max (t) represents the maximum value of the predicted value of the photovoltaic output at the tth hour of the day-ahead; and P pv_min (t) = 0.9P pv_pred (t), P pv_max (t) = 1.1P pv_pred (t);

[0037] The intra-day prediction model is used to output a corrected charge-discharge plan based on a 24-hour SOC reference trajectory, a benchmark charge-discharge plan and feedback data with a resolution of 15 minutes; the feedback data includes actual SOC, photovoltaic output deviation and frequency modulation response record;

[0038] The objective function output by the intra-day prediction model is to minimize the value of the prediction deviation function and is constrained by the photovoltaic output constraint function and the power fluctuation rate limiting function;

[0039] The formula of the prediction deviation function is:

[0040]

[0041] Min Cost = Min (λk ID represents the minimum value of the prediction deviation of the energy storage system; k is 0, 1, 2, …; λ ID (t) represents the predicted value of the electricity price at the tth hour of the day; P ref (t) represents the 24-hour SOC reference trajectory;

[0042] The formula of the photovoltaic output constraint function is:

[0043] P pv,ID (t) = P pv,DA (t) + ΔP pv,error (t) ;

[0044] P pv,ID (t) represents the photovoltaic output value at the tth hour of the day; P pv,DA (t) represents the photovoltaic prediction value at the tth hour of the day; ΔP pv,error (t) represents the photovoltaic prediction error value at the tth hour of the day, and ΔP pv,error (t) ~ N(0, σ 2 ), and σ represents the standard deviation of the normal distribution;

[0045] The formula of the power fluctuation rate limiting function is:

[0046] |P' grid (t) - P' grid (t-1) |≤ 0.1P' grid,max ;

[0047] P' grid (t) represents the grid power at the tth hour of the day; P' grid (t-1) represents the grid power at the (t-1)th hour of the day; P' grid,max represents the maximum grid output power of the day;

[0048] The real-time prediction model is used to output the charging and discharging instructions and the frequency modulation output power based on the modified charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output, and the actual SOC with a resolution of 1 minute;

[0049] The objective function output by the real-time prediction model is to maximize the value of the electricity charge income function, and is constrained by the real-time power balance function and the frequency modulation power limiting function;

[0050] The formula of the electricity charge income function is:

[0051]

[0052] where Max Profitt RT represents the maximum value of electricity revenue; λ RT (t) represents the real-time electricity price; P” dischar (t) represents the discharging power of the energy storage system at time t; P” char (t) represents the charging power of the energy storage system at time t; Δt RT represents the control period of the real-time prediction model; β represents the frequency regulation revenue weight coefficient; P FR (t) represents the power responding to the frequency regulation demand of the power grid at time t;

[0053] The formula of the real-time power balance function is:

[0054] P” grid (t) = P” load (t) + P” char (t) - P pv,actual (t) - P” dischar (t) + P FR (t);

[0055] where P” grid (t) represents the power grid power at time t; P” load (t) represents the load power at time t; P” char (t) represents the charging power of the energy storage system at time t; P pv,actual (t) represents the actual photovoltaic output at time t;

[0056] The formula of the frequency regulation power limitation function is:

[0057] 0 ≤ P FR (t) ≤ 0.3P rated ;

[0058] where P rated represents the rated power of the energy storage system.

[0059] Further, the step S6 is specifically:

[0060] The output of the day-ahead prediction model is optimized at a frequency of once every 24 hours based on a 24-hour sliding window; the output of the intra-day prediction model is optimized at a frequency of once every 15 minutes based on a 4-hour sliding window; and the output of the real-time prediction model is optimized at a frequency of once every 1 minute based on a 15-minute sliding window.

[0061] Further, in the step S7, the deviation compensation mechanism is specifically:

[0062] When the real-time photovoltaic output deviation is greater than 15% or the real-time electricity price fluctuation rate is greater than 20%, the charging power and discharging power of the energy storage system are adjusted based on preset priorities; based on the frequency modulation signal issued by the power grid, the energy storage system is switched to a frequency modulation mode, and the charging power and discharging power are adjusted according to P FR (t) = K·Δf(t) response frequency deviation; wherein, K represents a frequency modulation coefficient, and Δf(t) represents a power grid frequency offset.

[0063] In a second aspect, the present application provides a photovoltaic energy storage system based on multi-time scale rolling optimization, comprising the following modules:

[0064] A model creation module is configured to create and train an electricity price prediction model, a photovoltaic output prediction model, a load prediction model, a day-ahead prediction model, an intra-day prediction model, and a real-time prediction model.

[0065] A data acquisition module is configured to acquire historical electricity price data, historical photovoltaic output data, historical load data, and initial SOC of the energy storage system, input the historical electricity price data into the electricity price prediction model to obtain a 24-hour electricity price prediction curve, input the historical photovoltaic output data into the photovoltaic output prediction model to obtain a 24-hour photovoltaic output prediction curve, and input the historical load data into the load prediction model to obtain a 24-hour load prediction curve.

[0066] A day-ahead prediction module is configured to input the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve, and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference trajectory and a benchmark charging and discharging plan.

[0067] An intra-day prediction module is configured to acquire feedback data including actual SOC, photovoltaic output deviation, and frequency modulation response records, input the 24-hour SOC reference trajectory, the benchmark charging and discharging plan, and the feedback data into the intra-day prediction model to obtain a corrected charging and discharging plan.

[0068] A real-time prediction module is configured to set a dynamic SOC limit, acquire real-time electricity price and real-time photovoltaic output, input the corrected charging and discharging plan, the dynamic SOC limit, the real-time electricity price, the real-time photovoltaic output, and the actual SOC into the real-time prediction model to obtain charging and discharging instructions and frequency modulation output power, and perform energy storage regulation based on the charging and discharging instructions and the frequency modulation output power.

[0069] A rolling optimization module is configured to roll the outputs of the day-ahead prediction model, the intra-day prediction model, and the real-time prediction model.

[0070] A deviation compensation module is configured to set a deviation compensation mechanism and compensate the frequency modulation output power based on the deviation compensation mechanism.

[0071] Further, in the model creating module, the electricity price prediction model is constructed based on a first LSTM network, and is configured to output a 24-hour electricity price prediction curve according to historical electricity price data;

[0072] The photovoltaic output prediction model is constructed based on an NWP meteorological network and an XGBoost network, and is configured to output a 24-hour photovoltaic output prediction curve according to historical photovoltaic output data;

[0073] The load prediction model is constructed based on a second LSTM network, and is configured to output a 24-hour load prediction curve according to historical load data.

[0074] Further, in the model creating module, the day-ahead prediction model is configured to output a 24-hour SOC reference trajectory and a benchmark charging and discharging plan based on the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and an initial SOC, with an hour as a resolution; and the 24-hour SOC reference trajectory serves as a boundary condition of an intra-day prediction model.

[0075] The objective function output by the day-ahead prediction model is to maximize the value of a whole-cycle profit function, and is constrained by a storage dynamic constraint function, a grid interaction constraint function, a charging and discharging mutual exclusion constraint function and a photovoltaic output uncertainty robust optimization constraint function.

[0076] The formula of the whole-cycle profit function is:

[0077]

[0078] Wherein, Max Profit DA represents the maximum value of the whole-cycle profit of the energy storage system; λ DA (t) represents the electricity price prediction value of the tth hour of the day-ahead prediction; P dischar (t) represents the discharging power of the energy storage system in the tth hour of the day-ahead prediction; P char (t) represents the charging power of the energy storage system in the tth hour of the day-ahead prediction; Δt DA represents the time resolution of the day-ahead prediction model, i.e. 1 hour; C degradation represents the life loss cost based on the Rainflow counting method; C grid represents a grid interaction penalty term, which is configured to suppress power peak-valley fluctuation.

[0079] The formula of the storage dynamic constraint function is:

[0080]

[0081] SOC(t) = SOC(t-1) + η char charge efficiency of the energy storage system; and η rated discharge efficiency of the energy storage system; E grid,min representing the rated capacity of the energy storage system;

[0082] The formula of the grid interaction constraint function is:

[0083] P grid (t) = P load (t) + P char (t) - P pv_pred (t) - P dischar (t) ≤

[0084] P grid,max (t);

[0085] P grid,min (t) represents the minimum grid output power at the tth hour of the day-ahead; P grid (t) represents the grid power at the tth hour of the day-ahead; P load (t) represents the load power at the tth hour of the day-ahead; P pv_pred (t) represents the photovoltaic output prediction value at the tth hour of the day-ahead; P grid,max (t) represents the maximum grid output power at the tth hour of the day-ahead;

[0086] The formula of the charge-discharge mutual exclusion constraint function is:

[0087] P dischar (t) · P char (t) = 0;

[0088] The formula of the photovoltaic output uncertainty robust optimization constraint function is:

[0089] P pv_min (t) ≤ P pv_pred (t) ≤ P pv_max (t);

[0090] P pv_min (t) represents the minimum value of the photovoltaic output prediction value at the tth hour of the day-ahead; P pv_max (t) represents the maximum value of the photovoltaic output prediction value at the tth hour of the day-ahead; and P pv_min (t) = 0.9P pv_pred (t), P pv_max (t) = 1.1P pv_pred (t);

[0091] The intraday prediction model is used to output a revised charge and discharge plan with a 15-minute resolution based on a 24-hour SOC reference trajectory, a baseline charge and discharge plan, and feedback data; the feedback data includes actual SOC, photovoltaic output deviation, and frequency modulation response records;

[0092] The objective function output by the intraday forecast model is to minimize the value of the forecast deviation function, and is constrained by the photovoltaic output constraint function and the power fluctuation rate limit function;

[0093] The formula of the prediction deviation function is:

[0094]

[0095] Among them, Min Cost ID Indicates the minimum value of the energy storage system prediction deviation; the value of k is 0, 1, 2...; λ ID (t) represents the electricity price forecast value at the tth hour of the day; represents the discharge power of the energy storage system at the tth hour of the day; α represents the SOC tracking weight coefficient, which is used to control the penalty for deviations from the 24-hour SOC reference trajectory and the benchmark charge and discharge plan; SOC'(t) represents the SOC of the energy storage system at the tth hour of the day; SOC' ref (t) represents the 24-hour SOC reference trajectory;

[0096] The formula of the photovoltaic output constraint function is:

[0097] P pv,ID (t) = P pv,DA (t)+ΔP pv,error (t);

[0098] Among them, P pv,ID (t) represents the photovoltaic output value at time t within the day; P pv,DA (t) represents the photovoltaic forecast value at time t the day before; ΔP pv,error (t) represents the photovoltaic prediction error value at time t, and ΔP pv,error (t)~N(0,σ 2 ), σ represents the standard deviation of the normal distribution;

[0099] The formula of the power fluctuation rate limiting function is:

[0100] |P' grid (t)-P' grid (t-1)|≤0.1P' grid,max ;

[0101] Among them, P' grid (t) represents the power of the power grid at time t within the day; P' grid(t-1) represents the grid power at time t-1 within a day; P' grid,max represents the maximum grid output within a day;

[0102] The real-time prediction model is used to output the charging and discharging instructions and the frequency modulation output power based on the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC with a resolution of 1 minute;

[0103] The objective function output by the real-time prediction model is to maximize the value of the electricity charge income function, and is constrained by the real-time power balance function and the frequency modulation power limit function;

[0104] The formula of the electricity charge income function is:

[0105]

[0106] Wherein, Max Profitt RT represents the maximum value of the electricity charge income; λ RT (t) represents the real-time electricity price; P" dischar (t) represents the discharging power of the energy storage system at time t; P" char (t) represents the charging power of the energy storage system at time t; Δt RT represents the control period of the real-time prediction model; β represents the frequency modulation income weight coefficient; P FR (t) represents the power responding to the grid frequency modulation demand at time t;

[0107] The formula of the real-time power balance function is:

[0108] P" grid (t) = P" load (t) + P" char (t) - P pv,actual (t) - P" dischar (t) + P FR (t);

[0109] Wherein, P" grid (t) represents the grid power at time t; P" load (t) represents the load power at time t; P" char (t) represents the charging power of the energy storage system at time t; P pv,actual (t) represents the actual photovoltaic output at time t;

[0110] The formula of the frequency modulation power limit function is:

[0111] 0 ≤ P FR (t) ≤ 0.3P rated ;

[0112] Wherein, Prated represents the rated power of the energy storage system.

[0113] Further, the rolling optimization module is specifically used for:

[0114] The output of the day-ahead prediction model is rolling optimized at a frequency of once per 24 hours based on a 24-hour sliding window; the output of the intra-day prediction model is rolling optimized at a frequency of once per 15 minutes based on a 4-hour sliding window; and the output of the real-time prediction model is rolling optimized at a frequency of once per 1 minute based on a 15-minute sliding window.

[0115] Further, in the deviation compensation module, the deviation compensation mechanism is specifically:

[0116] When the real-time photovoltaic output deviation is greater than 15% or the real-time electricity price fluctuation rate is greater than 20%, the charging power and the discharging power of the energy storage system are adjusted based on a preset priority; based on the frequency modulation signal issued by the power grid, the energy storage system is switched to a frequency modulation mode, and the energy storage system is controlled to operate in the frequency modulation mode according to P FR (t) = K·Δf(t) response frequency deviation; wherein K represents a frequency modulation coefficient, and Δf(t) represents a power grid frequency offset.

[0117] The present application has the advantages of:

[0118] By creating and training the electricity price prediction model, the photovoltaic output prediction model, the load prediction model, the day-ahead prediction model, the intra-day prediction model and the real-time prediction model; then inputting the electricity price historical data into the electricity price prediction model to obtain the 24-hour electricity price prediction curve, inputting the photovoltaic output historical data into the photovoltaic output prediction model to obtain the 24-hour photovoltaic output prediction curve, and inputting the load historical data into the load prediction model to obtain the 24-hour load prediction curve; inputting the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and the initial SOC into the day-ahead prediction model to obtain the 24-hour SOC reference trajectory and the benchmark charging and discharging plan; then collecting the feedback data including the actual SOC, the photovoltaic output deviation and the frequency modulation response record, inputting the 24-hour SOC reference trajectory, the benchmark charging and discharging plan and the feedback data into the intra-day prediction model to obtain the corrected charging and discharging plan; then setting the SOC dynamic limit value, collecting the real-time electricity price and the real-time photovoltaic output, inputting the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC into the real-time prediction model to obtain the charging and discharging instruction and the frequency modulation output power, and performing energy storage regulation and control based on the charging and discharging instruction and the frequency modulation output power; rolling optimization of the output of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model, and compensating the frequency modulation output power based on the set deviation compensation mechanism; that is, the energy storage regulation and control is performed based on the multi-time scale (24 hours of day-ahead, 4 hours of intra-day, and 15 minutes of real-time) of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model, so as to overcome the problem of single time scale in the traditional method; the charging power and the discharging power of the energy storage system are dynamically adjusted through the deviation compensation mechanism, so as to overcome the problem of cumulative prediction error in the traditional method, avoid frequent adjustment of the charging and discharging of the energy storage system, and participate in frequency modulation to adapt to the power grid; by setting the constraint function of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model, the battery life loss cost is fully considered, the short-term benefit is avoided to be excessively pursued, and the full life cycle economy of the energy storage system is sacrificed, and finally the accuracy, the economy and the power grid adaptability of the energy storage regulation and control are greatly improved, and the battery life attenuation speed is greatly slowed down. BRIEF DESCRIPTION OF DRAWINGS

[0119] The application will be further described below with reference to the drawings and embodiments.

[0120] Fig. 1 is a flowchart of a photovoltaic energy storage regulation and control method based on multi-time scale rolling optimization.

[0121] Fig. 2 is a structural schematic diagram of a photovoltaic energy storage regulation and control system based on multi-time scale rolling optimization. DETAILED DESCRIPTION

[0122] The technical solutions in the embodiments of the present application have the following general idea: based on multi-time scales (24 hours in advance, 4 hours in advance, and 15 minutes in real time) of a day-ahead prediction model, an intra-day prediction model, and a real-time prediction model, energy storage regulation is performed, and the problem of using a single time scale in the prior art is overcome; through a deviation compensation mechanism, the charging power and discharging power of the energy storage system are dynamically adjusted to overcome the problem of cumulative prediction error in the prior art, avoid frequent adjustment of the charging and discharging of the energy storage system, and participate in frequency regulation to adapt to the power grid; by setting constraint functions of the day-ahead prediction model, the intra-day prediction model, and the real-time prediction model, the battery life loss cost is fully considered, and the short-term benefit is avoided to sacrifice the economy of the whole life cycle of the energy storage system, thereby improving the accuracy, economy, and power grid adaptability of the energy storage regulation, and delaying the battery life attenuation speed.

[0123] Please refer to Figs. 1-2 The preferred embodiment of the photovoltaic energy storage regulation method based on multi-time scale rolling optimization of the present application includes the following steps:

[0124] Step S1, a server creates and trains an electricity price prediction model, a photovoltaic output prediction model, a load prediction model, a day-ahead prediction model, an intra-day prediction model, and a real-time prediction model; in specific implementation, the output priority of the real-time prediction model is higher than that of the day-ahead prediction model and the intra-day prediction model, so as to ensure the emergency working condition response;

[0125] Step S2, the server acquires electricity price historical data, photovoltaic output historical data, load historical data, and the initial SOC of the energy storage system, inputs the electricity price historical data into the electricity price prediction model to obtain a 24-hour electricity price prediction curve, inputs the photovoltaic output historical data into the photovoltaic output prediction model to obtain a 24-hour photovoltaic output prediction curve, and inputs the load historical data into the load prediction model to obtain a 24-hour load prediction curve;

[0126] Step S3, the server inputs the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve, and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference track and a benchmark charging and discharging plan;

[0127] Step S4, the server collects feedback data including the actual SOC, the photovoltaic output deviation, and the frequency regulation response record, inputs the 24-hour SOC reference track, the benchmark charging and discharging plan, and the feedback data into the intra-day prediction model to obtain a corrected charging and discharging plan;

[0128] Step S5, the server sets the SOC dynamic limit value, collects the real-time electricity price and the real-time photovoltaic output, inputs the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC into a real-time prediction model to obtain a charging and discharging instruction and a frequency modulation output power, and performs energy storage regulation and control based on the charging and discharging instruction and the frequency modulation output power; a regulation and control log is generated based on the charging and discharging instruction and the frequency modulation output power, the regulation and control log is encrypted into an encrypted log, and the encrypted log is stored and distributedly backed up; the SOC dynamic limit value is [SOC min ,SOC max ];

[0129] The encryption process of the regulation and control log is specifically: obtaining a current timestamp, performing hash calculation on the regulation and control log and the timestamp to obtain a hash value, encrypting the regulation and control log, the timestamp and the hash value through an AES algorithm to obtain first-level encrypted data, obtaining second-level encrypted data by cyclically shifting each character of the first-level encrypted data to the right by three bits, and encrypting the second-level encrypted data into an encrypted log through an RC6 algorithm; by combining the timestamp, the hash value, the AES algorithm, the RC6 algorithm, the bit shift direction and the bit shift number, six security measures are taken, which greatly improves the security of the regulation and control log storage backup, avoids being stolen and tampered with in plaintext, and is convenient for later tracing;

[0130] Step S6, the server rolls the output of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model;

[0131] Step S7, the server sets a deviation compensation mechanism, and compensates the frequency modulation output power based on the deviation compensation mechanism.

[0132] The application effectively improves the economic benefits of the energy storage system under the real-time electricity price fluctuation and the intermittent photovoltaic scene, reduces the influence of the prediction error on the energy storage regulation and control, prolongs the service life of the battery, and enhances the collaborative ability of the energy storage system and the power grid by using the multi-time scale rolling optimization, the constraint function combined with the battery life benefit joint optimization and the deviation compensation mechanism (dynamic response mechanism).

[0133] In the step S1, the electricity price prediction model is constructed based on a first LSTM network, and is used to output a 24-hour electricity price prediction curve according to electricity price historical data;

[0134] The photovoltaic output prediction model is constructed based on an NWP meteorological network and an XGBoost network, and is used to output a 24-hour photovoltaic output prediction curve according to photovoltaic output historical data;

[0135] The load prediction model is constructed based on a second LSTM network, and is used to output a 24-hour load prediction curve according to load historical data.

[0136] In the step S1, the day-ahead prediction model is used to output a 24-hour SOC reference trajectory and a benchmark charging and discharging plan based on a 24-hour electricity price prediction curve, a 24-hour photovoltaic output prediction curve, a 24-hour load prediction curve and an initial SOC with a resolution of 1 hour; the 24-hour SOC reference trajectory serves as a boundary condition of the intra-day prediction model;

[0137] The objective function output by the day-ahead prediction model is to maximize the value of the whole-cycle profit function, and is constrained by a storage dynamic constraint function, a grid interaction constraint function, a charging and discharging mutual exclusion constraint function and a photovoltaic output uncertainty robust optimization constraint function;

[0138] The formula of the whole-cycle profit function is:

[0139]

[0140] Wherein, Max Profit DA represents the maximum value of the whole-cycle profit of the energy storage system; λ DA (t) represents the electricity price prediction value of the tth hour of the day-ahead; P dischar (t) represents the discharging power of the energy storage system in the tth hour of the day-ahead; P char (t) represents the charging power of the energy storage system in the tth hour of the day-ahead; Δt DA represents the time resolution of the day-ahead prediction model, i.e. 1 hour; C degradation represents the life loss cost based on the Rainflow counting method; C grid represents the grid interaction penalty term, which is used to suppress the power peak-valley fluctuation;

[0141]

[0142] C grid =Σ[μ peak ·max(P grid (t))+μ valley ·min(P grid (t))];

[0143] Wherein, k represents the unit cycle cost of the battery; DoD t represents the tth charging and discharging depth; n represents the fitting coefficient, and the value range is (1.2, 1.5); μ peak and μ valley both represent the grid peak-valley power penalty coefficient;

[0144] The formula of the storage dynamic constraint function is:

[0145]

[0146] SOC(t) = SOC(t - 1) + η char charge efficiency of the energy storage system; and η rated discharge efficiency of the energy storage system; and E grid,min rated capacity of the energy storage system;

[0147] The formula of the grid interaction constraint function is:

[0148] P grid (t) = P load (t) + P char (t) - P pv_pred (t) - P dischar (t) ≤

[0149] P grid,max (t);

[0150] P grid,min (t) represents the minimum grid output power at the tth hour of the day-ahead; P grid (t) represents the grid power at the tth hour of the day-ahead; P load (t) represents the load power at the tth hour of the day-ahead; P pv_pred (t) represents the photovoltaic output prediction value at the tth hour of the day-ahead; P grid,max (t) represents the maximum grid output power at the tth hour of the day-ahead;

[0151] The formula of the charge-discharge mutual exclusion constraint function is:

[0152] P dischar (t) · P char (t) = 0;

[0153] The formula of the photovoltaic output uncertainty robust optimization constraint function is:

[0154] P pv_min (t) ≤ P pv_pred (t) ≤ P pv_max (t);

[0155] P pv_min (t) represents the minimum value of the photovoltaic output prediction value at the tth hour of the day-ahead; P pv_max (t) represents the maximum value of the photovoltaic output prediction value at the tth hour of the day-ahead; and P pv_min (t) = 0.9P pv_pred (t), P pv_max (t) = 1.1P pv_pred (t), that is, covering 90% of the prediction confidence interval;

[0156] The intraday prediction model is used to output a corrected charge-discharge plan based on a 24-hour SOC reference track, a benchmark charge-discharge plan and feedback data with a resolution of 15 minutes; the feedback data includes actual SOC, photovoltaic output deviation and frequency modulation response record; the corrected charge-discharge plan serves as a boundary condition of a real-time prediction model;

[0157] The objective function output by the intraday prediction model is to minimize the value of a prediction deviation function and is constrained by a photovoltaic output constraint function and a power fluctuation rate limiting function;

[0158] The formula of the prediction deviation function is:

[0159]

[0160] Wherein, Min Cost ID represents the minimum value of the prediction deviation of the energy storage system; the value of k is 0, 1, 2, …; λ ID (t) represents the predicted value of the electricity price at the tth hour of the day; P'(t) represents the discharging power of the energy storage system at the tth hour of the day; α represents an SOC tracking weight coefficient for controlling the deviation penalty of the 24-hour SOC reference track from the benchmark charge-discharge plan; SOC'(t) represents the SOC of the energy storage system at the tth hour of the day; SOC' ref (t) represents the 24-hour SOC reference track;

[0161] The formula of the photovoltaic output constraint function is:

[0162] P pv,ID (t) = P pv,DA (t) + ΔP pv,error (t);

[0163] Wherein, P pv,ID (t) represents the photovoltaic output value at the tth hour of the day; P pv,DA (t) represents the photovoltaic prediction value at the tth hour of the day; ΔP pv,error (t) represents the photovoltaic prediction error value at the tth hour of the day, and ΔP pv,error (t) ~ N(0, σ 2 ), and σ represents the standard deviation of the normal distribution;

[0164] The formula of the power fluctuation rate limiting function is:

[0165] |P' grid (t) - P' grid (t-1)| ≤ 0.1P' grid,max ;

[0166] Wherein, P' grid (t) represents the grid power at the tth hour of the day; P'grid (t-1) represents the grid power at time t-1 within a day; P' grid,max represents the maximum grid output within a day;

[0167] The real-time prediction model is used to output the charging and discharging instructions and the frequency modulation output power based on the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC with a resolution of 1 minute;

[0168] The objective function output by the real-time prediction model is to maximize the value of the electricity charge income function, and is constrained by the real-time power balance function and the frequency modulation power limit function;

[0169] The formula of the electricity charge income function is:

[0170]

[0171] Wherein, Max Profitt RT represents the maximum value of the electricity charge income; λ RT (t) represents the real-time electricity price; P" dischar (t) represents the discharging power of the energy storage system at time t; P" char (t) represents the charging power of the energy storage system at time t; Δt RT represents the control period of the real-time prediction model; β represents the frequency modulation income weight coefficient; P FR (t) represents the power responding to the grid frequency modulation demand at time t;

[0172] The formula of the real-time power balance function is:

[0173] P" grid (t) = P" load (t) + P" char (t) - P pv,actual (t) - P" dischar (t) + P FR (t);

[0174] Wherein, P" grid (t) represents the grid power at time t; P" load (t) represents the load power at time t; P" char (t) represents the charging power of the energy storage system at time t; P pv,actual (t) represents the actual photovoltaic output at time t;

[0175] The formula of the frequency modulation power limit function is:

[0176] 0≤P FR (t)≤0.3P rated ;

[0177] where P rated represents the rated power of the energy storage system.

[0178] The day-ahead prediction model, the intra-day prediction model, and the real-time prediction model of the present application are further described as follows:

[0179] Day-ahead prediction model (24-hour rolling window): Based on the price prediction and the photovoltaic output prediction, a benchmark charging and discharging plan is generated with a resolution of 1 hour.

[0180] Intra-day prediction model (4-hour rolling window): The prediction data is updated every 4 hours, and the benchmark charging and discharging plan is corrected using a sliding window mechanism to address short-term fluctuations in photovoltaic output and price prediction deviations.

[0181] Real-time prediction model (15-minute rolling window): Real-time prices (power market API interface) and actual photovoltaic output of the power grid are collected every 15 minutes, and the energy storage power is dynamically adjusted to compensate for prediction errors and respond to grid frequency regulation requirements.

[0182] The step S6 specifically includes:

[0183] The output of the day-ahead prediction model is optimized based on a 24-hour sliding window with a frequency of once every 24 hours; the output of the intra-day prediction model is optimized based on a 4-hour sliding window with a frequency of once every 15 minutes; and the output of the real-time prediction model is optimized based on a 15-minute sliding window with a frequency of once every 1 minute.

[0184] That is, a rolling optimization closed loop is constructed for each time level (day-ahead, intra-day, and real-time), and the steps are:

[0185] a. Prediction time domain division

[0186] Day-ahead: 24-hour sliding window, global optimization once every 24 hours.

[0187] Intra-day: 4-hour sliding window, rolling once every 15 minutes, covering fine adjustment for the next 4 hours.

[0188] Real-time: 15-minute sliding window, rolling once every 1 minute, achieving second-level response.

[0189] b. Rolling execution steps

[0190] ① Initialization: Obtain the latest price, photovoltaic output, and SOC state.

[0191] ② Rolling optimization: Call the prediction model to update the future window data, solve the objective function based on the current state, and generate the control sequence.

[0192] ③ Instruction execution: only the first time power instruction is executed, and the rest of the period prediction value is taken as a reference.

[0193] ④ Feedback update: sliding window to the next time, repeat steps ①-③.

[0194] c. Time window parameters

[0195]

[0196] In the step S7, the deviation compensation mechanism is specifically:

[0197] When the real-time photovoltaic output deviation is greater than 15%, or the real-time electricity price fluctuation rate is greater than 20%, the charging power and discharging power of the energy storage system are adjusted based on a preset priority; based on the frequency modulation signal issued by the power grid, the energy storage system is switched to the frequency modulation mode, and the P FR (t) = K·Δf(t) response frequency deviation; wherein, K represents a frequency modulation coefficient, and Δf(t) represents a power grid frequency offset.

[0198] The preferred embodiment of the photovoltaic energy storage system based on multi-time scale rolling optimization comprises the following modules:

[0199] The model creation module is used for the server to create and train an electricity price prediction model, a photovoltaic output prediction model, a load prediction model, a day-ahead prediction model, a day-ahead prediction model, and a real-time prediction model; in specific implementation, the priority of the output of the real-time prediction model is higher than that of the day-ahead prediction model and the day-ahead prediction model, so as to ensure the emergency working condition response;

[0200] The data acquisition module is used for the server to acquire electricity price historical data, photovoltaic output historical data, load historical data, and initial SOC of the energy storage system, input the electricity price historical data into the electricity price prediction model to obtain a 24-hour electricity price prediction curve, input the photovoltaic output historical data into the photovoltaic output prediction model to obtain a 24-hour photovoltaic output prediction curve, and input the load historical data into the load prediction model to obtain a 24-hour load prediction curve.

[0201] The day-ahead prediction module is used for the server to input the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve, and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference track and a benchmark charging and discharging plan.

[0202] The day-ahead prediction module is used for the server to input the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve, and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference track and a benchmark charging and discharging plan.

[0203] The real-time prediction module is used for setting a SOC dynamic limit value by the server, collecting real-time electricity price and real-time photovoltaic output, inputting the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC into a real-time prediction model to obtain a charging and discharging instruction and a frequency modulation output power, and performing energy storage regulation and control based on the charging and discharging instruction and the frequency modulation output power; a regulation and control log is generated based on the charging and discharging instruction and the frequency modulation output power, the regulation and control log is encrypted into an encrypted log, and the encrypted log is stored and distributedly backed up; the SOC dynamic limit value is that the value of the SOC falls within [SOC min , SOC max ];

[0204] The encryption process of the regulation and control log is specifically that a current timestamp is obtained, hash calculation is performed on the regulation and control log and the timestamp to obtain a hash value, the regulation and control log, the timestamp and the hash value are encrypted by an AES algorithm to obtain first-level encrypted data, each character of the first-level encrypted data is cyclically shifted right by three bits to obtain second-level encrypted data, and the second-level encrypted data is encrypted into the encrypted log by an RC6 algorithm; by combining the timestamp, the hash value, the AES algorithm, the RC6 algorithm, a bit shift direction and a bit shift number, six security measures are taken, the security of regulation and control log storage and backup is greatly improved, the regulation and control log is prevented from being stolen and tampered with in plaintext, and traceability is facilitated in the later period.

[0205] The rolling optimization module is used for rolling optimization of the output of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model by the server.

[0206] The deviation compensation module is used for setting a deviation compensation mechanism by the server, and compensating the frequency modulation output power based on the deviation compensation mechanism.

[0207] The application effectively improves the economic benefits of the energy storage system under real-time electricity price fluctuation and intermittent photovoltaic scenes, reduces the influence of prediction error on energy storage regulation and control, prolongs the service life of the battery, and enhances the collaborative ability of the energy storage system and the power grid by means of multi-time scale rolling optimization, a constraint function combined with the benefits of battery life and a deviation compensation mechanism (dynamic response mechanism).

[0208] In the model creation module, the electricity price prediction model is constructed based on a first LSTM network and is used for outputting a 24-hour electricity price prediction curve according to electricity price historical data;

[0209] The photovoltaic output prediction model is constructed based on an NWP meteorological network and an XGBoost network and is used for outputting a 24-hour photovoltaic output prediction curve according to photovoltaic output historical data;

[0210] The load prediction model is constructed based on a second LSTM network and is used for outputting a 24-hour load prediction curve according to load historical data.

[0211] In the model creation module, the day-ahead prediction model is used to output a 24-hour SOC reference trajectory and a benchmark charging and discharging plan based on a 24-hour electricity price prediction curve, a 24-hour photovoltaic output prediction curve, a 24-hour load prediction curve and an initial SOC with an hour as a resolution; the 24-hour SOC reference trajectory serves as a boundary condition of an intra-day prediction model;

[0212] The objective function output by the day-ahead prediction model is to maximize the value of a full-cycle profit function, and is constrained by a storage dynamic constraint function, a grid interaction constraint function, a charging and discharging mutual exclusion constraint function and a photovoltaic output uncertainty robust optimization constraint function;

[0213] The formula of the full-cycle profit function is:

[0214]

[0215] Wherein, Max Profit DA represents the maximum value of the full-cycle profit of the energy storage system; λ DA (t) represents the electricity price prediction value of the tth hour of the day-ahead; P dischar (t) represents the discharging power of the energy storage system in the tth hour of the day-ahead; P char (t) represents the charging power of the energy storage system in the tth hour of the day-ahead; Δt DA represents the time resolution of the day-ahead prediction model, i.e. 1 hour; C degradation represents the life loss cost based on the Rainflow counting method; C grid represents a grid interaction penalty term for suppressing power peak-valley fluctuation;

[0216]

[0217] C grid =∑[μ peak ·max(P grid (t))+μ valley ·min(P grid (t))];

[0218] Wherein, k represents the unit cycle cost of the battery; DoD t represents the tth charging and discharging depth; n represents a fitting coefficient, and the value range is (1.2, 1.5); μ peak and μ valley both represent grid peak-valley power penalty coefficients;

[0219] The formula of the storage dynamic constraint function is:

[0220]

[0221] Where SOC(t) represents the SOC of the energy storage system at the tth hour before the day; SOC(t-1) represents the SOC of the energy storage system at the t-1th hour before the day; η char Indicates the charging efficiency of the energy storage system; Indicates the discharging efficiency of the energy storage system; E rated Indicates the rated capacity of the energy storage system;

[0222] The formula of the grid interaction constraint function is:

[0223] P grid,min (t)≤P grid (t) = P load (t)+P char (t)-P pv_pred (t)-P dischar (t)≤

[0224] P grid,max (t);

[0225] Among them, P grid,min (t) represents the minimum power output of the power grid at hour t before the day; P grid (t) represents the power grid power at hour t before the day; P load (t) represents the load power at hour t before the current day; P pv_pred (t) represents the photovoltaic output forecast value at the tth hour before the day; P grid,max (t) represents the maximum power output of the power grid at hour t before the day;

[0226] The formula of the charge-discharge mutual exclusion constraint function is:

[0227] P dischar (t)·P char (t) = 0;

[0228] The formula of the photovoltaic output uncertainty robust optimization constraint function is:

[0229] P pv_min (t)≤P pv_pred (t)≤P pv_max (t);

[0230] Among them, P pv_min (t) represents the minimum value of the photovoltaic output forecast value at the tth hour before the day; P pv_max (t) represents the maximum value of the photovoltaic output forecast value at hour t before the day; and P pv_min (t)=0.9P pv_pred (t), P pv_max (t)=1.1P pv_pred (t), i.e., covering the 90% prediction confidence interval;

[0231] The intraday prediction model is used to output a corrected charge-discharge plan based on a 24-hour SOC reference track, a benchmark charge-discharge plan and feedback data with a resolution of 15 minutes; the feedback data includes actual SOC, photovoltaic output deviation and frequency modulation response record; the corrected charge-discharge plan serves as a boundary condition of a real-time prediction model;

[0232] The objective function output by the intraday prediction model is to minimize the value of a prediction deviation function and is constrained by a photovoltaic output constraint function and a power fluctuation rate limiting function;

[0233] The formula of the prediction deviation function is:

[0234]

[0235] Wherein, Min Cost ID represents the minimum value of the prediction deviation of the energy storage system; the value of k is 0, 1, 2, …; λ ID (t) represents the predicted value of the electricity price at the tth hour of the day; P'(t) represents the discharging power of the energy storage system at the tth hour of the day; α represents an SOC tracking weight coefficient for controlling the deviation penalty of the 24-hour SOC reference track from the benchmark charge-discharge plan; SOC'(t) represents the SOC of the energy storage system at the tth hour of the day; SOC'(t) represents the SOC of the energy storage system at the tth hour of the day; and SOC'(t) represents the SOC of the energy storage system at the tth hour of the day. ref (t) represents the 24-hour SOC reference track.

[0236] The formula of the photovoltaic output constraint function is:

[0237] P pv,ID (t)=P pv,DA (t)+ΔP pv,error (t);

[0238] Wherein, P pv,ID (t) represents the photovoltaic output value at the tth hour of the day; P pv,DA (t) represents the photovoltaic prediction value at the tth hour of the day; ΔP pv,error (t) represents the photovoltaic prediction error value at the tth hour of the day, and ΔP pv,error (t) ~ N(0, σ 2 ), and σ represents the standard deviation of the normal distribution.

[0239] The formula of the power fluctuation rate limiting function is:

[0240] |P' grid (t)-P' grid (t-1)|≤0.1P' grid,max ;

[0241] Wherein, P' grid (t) represents the grid power at the tth hour of the day; P'grid (t-1) represents the grid power at time t-1 within a day; P' grid,max represents the maximum grid output within a day;

[0242] The real-time prediction model is used to output the charging and discharging instructions and the frequency modulation output power based on the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC with a resolution of 1 minute;

[0243] The objective function output by the real-time prediction model is to maximize the value of the electricity charge income function, and is constrained by the real-time power balance function and the frequency modulation power limit function;

[0244] The formula of the electricity charge income function is:

[0245]

[0246] Wherein, Max Profitt RT represents the maximum value of the electricity charge income; λ RT (t) represents the real-time electricity price; P" dischar (t) represents the discharging power of the energy storage system at time t; P" char (t) represents the charging power of the energy storage system at time t; Δt RT represents the control period of the real-time prediction model; β represents the frequency modulation income weight coefficient; P FR (t) represents the power responding to the grid frequency modulation demand at time t;

[0247] The formula of the real-time power balance function is:

[0248] P" grid (t) = P" load (t) + P" char (t) - P pv,actual (t) - P" dischar (t) + P FR (t);

[0249] Wherein, P" grid (t) represents the grid power at time t; P" load (t) represents the load power at time t; P" char (t) represents the charging power of the energy storage system at time t; P pv,actual (t) represents the actual photovoltaic output at time t;

[0250] The formula of the frequency modulation power limit function is:

[0251] 0≤P FR (t)≤0.3P rated ;

[0252] P rated denotes the rated power of the energy storage system.

[0253] The day-ahead prediction model, the intra-day prediction model, and the real-time prediction model of the present application are further described as follows:

[0254] Day-ahead prediction model (24-hour rolling window): Based on the price prediction and the photovoltaic output prediction, a benchmark charging and discharging plan is generated with a resolution of 1 hour.

[0255] Intra-day prediction model (4-hour rolling window): The prediction data is updated every 4 hours, and the benchmark charging and discharging plan is corrected using a sliding window mechanism to address short-term fluctuations in photovoltaic output and price prediction deviations.

[0256] Real-time prediction model (15-minute rolling window): Real-time prices (power market API interface) and actual photovoltaic output of the power grid are collected every 15 minutes, and the energy storage power is dynamically adjusted to compensate for prediction errors and respond to grid frequency regulation requirements.

[0257] The rolling optimization module is specifically used for:

[0258] Based on a 24-hour sliding window, the output of the day-ahead prediction model is optimized every 24 hours; based on a 4-hour sliding window, the output of the intra-day prediction model is optimized every 15 minutes; and based on a 15-minute sliding window, the output of the real-time prediction model is optimized every 1 minute.

[0259] That is, a rolling optimization closed loop is constructed for each time level (day-ahead, intra-day, and real-time), and the steps are as follows:

[0260] a. Prediction time domain division

[0261] Day-ahead: 24-hour sliding window, global optimization once every 24 hours.

[0262] Intra-day: 4-hour sliding window, rolling every 15 minutes, covering fine adjustment for the next 4 hours.

[0263] Real-time: 15-minute sliding window, rolling every 1 minute, achieving second-level response.

[0264] b. Rolling execution steps

[0265] ① Initialization: Obtain the latest price, photovoltaic output, and SOC state.

[0266] ② Rolling optimization: Call the prediction model to update the future window data, solve the objective function based on the current state, and generate the control sequence.

[0267] III. Instruction execution: only the first time power instruction is executed, and the rest of the period prediction value is taken as a reference.

[0268] IV. Feedback update: sliding window to the next time, repeat steps I-III.

[0269] c. Time window parameters

[0270]

[0271] In the deviation compensation module, the deviation compensation mechanism is specifically:

[0272] When the real-time photovoltaic output deviation is greater than 15%, or the real-time electricity price fluctuation rate is greater than 20%, the charging power and discharging power of the energy storage system are adjusted based on the preset priority; based on the frequency modulation signal issued by the power grid, the energy storage system is switched to the frequency modulation mode, and the P FR (t) = K·Δf(t) response frequency deviation; wherein, K represents the frequency modulation coefficient, and Δf(t) represents the power grid frequency offset.

[0273] In summary, the advantages of the present application are:

[0274] The 24-hour SOC reference trajectory and the benchmark charge-discharge plan are obtained by inputting the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and the initial SOC into the day-ahead prediction model; then the feedback data including the actual SOC, the photovoltaic output deviation and the frequency modulation response record are collected, the 24-hour SOC reference trajectory, the benchmark charge-discharge plan and the feedback data are inputted into the day-in prediction model to obtain the corrected charge-discharge plan; then the SOC dynamic limit value is set, the real-time electricity price and the real-time photovoltaic output are collected, the corrected charge-discharge plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and the actual SOC are inputted into the real-time prediction model to obtain the charge-discharge instruction and the frequency modulation output power, and the energy storage is regulated and controlled based on the charge-discharge instruction and the frequency modulation output power; the output of the day-ahead prediction model, the day-in prediction model and the real-time prediction model is rolled and optimized, and the frequency modulation output power is compensated based on the set deviation compensation mechanism; that is, the energy storage is regulated and controlled based on the multi-time scale (24 hours of day-ahead, 4 hours of day-in and 15 minutes of real-time) of the day-ahead prediction model, the day-in prediction model and the real-time prediction model, so as to overcome the problem of single time scale in the prior art; the charging power and the discharging power of the energy storage system are dynamically adjusted through the deviation compensation mechanism, so as to overcome the problem of cumulative prediction error in the prior art, avoid frequent adjustment of the charge-discharge of the energy storage system and participate in frequency modulation to adapt to the power grid; the constraint function of the day-ahead prediction model, the day-in prediction model and the real-time prediction model is set, the battery life loss cost is fully considered, the short-term benefit is avoided to be excessively pursued and the whole life cycle economy of the energy storage system is sacrificed, the accuracy, the economy and the power grid adaptability of the energy storage regulation and control are greatly improved, and the battery life attenuation speed is greatly delayed.

[0275] Although the specific embodiments of the present application are described above, those skilled in the art should understand that the specific embodiments described are only illustrative, and are not intended to limit the scope of the present application, and equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present application should be covered within the scope of the claims of the present application.

Claims

1. A method for optical storage energy regulation based on multi-time scale rolling optimization, characterized in that: The method comprises the following steps: Step S1, creating and training an electricity price prediction model, a photovoltaic output prediction model, a load prediction model, a day-ahead prediction model, an intra-day prediction model, and a real-time prediction model; Step S2, obtaining electricity price historical data, photovoltaic output historical data, load historical data, and an initial SOC of the energy storage system, inputting the electricity price historical data into the electricity price prediction model to obtain a 24-hour electricity price prediction curve, inputting the photovoltaic output historical data into the photovoltaic output prediction model to obtain a 24-hour photovoltaic output prediction curve, and inputting the load historical data into the load prediction model to obtain a 24-hour load prediction curve; Step S3, inputting the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve, and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference trajectory and a benchmark charging and discharging plan; Step S4, collecting feedback data including an actual SOC, a photovoltaic output deviation, and a frequency modulation response record, inputting the 24-hour SOC reference trajectory, the benchmark charging and discharging plan, and the feedback data into the intra-day prediction model to obtain a corrected charging and discharging plan; Step S5, setting a dynamic SOC limit value, collecting real-time electricity price and real-time photovoltaic output, inputting the corrected charging and discharging plan, the dynamic SOC limit value, the real-time electricity price, the real-time photovoltaic output, and the actual SOC into the real-time prediction model to obtain charging and discharging instructions and frequency modulation output power, and performing energy storage regulation and control based on the charging and discharging instructions and the frequency modulation output power; Step S6, rolling optimization of the output of the day-ahead prediction model, the intra-day prediction model, and the real-time prediction model; Step S7, setting a deviation compensation mechanism, and compensating the frequency modulation output power based on the deviation compensation mechanism; the deviation compensation mechanism is specifically: When the real-time photovoltaic output deviation is greater than 15% or the real-time electricity price fluctuation rate is greater than 20%, the charging power and discharging power of the energy storage system are adjusted based on preset priorities; based on the frequency modulation signal issued by the power grid, the energy storage system is switched to a frequency modulation mode, and the energy storage system is charged or discharged according to P FR (t) = K·Δf(t) response frequency deviation; wherein K represents a frequency modulation coefficient, and Δf(t) represents a power grid frequency offset. 2.The energy storage regulation method based on multi-time scale rolling optimization of claim 1, wherein: In the step S1, the electricity price prediction model is constructed based on a first LSTM network, and is used to output a 24-hour electricity price prediction curve according to electricity price historical data; The photovoltaic output prediction model is constructed based on an NWP meteorological network and an XGBoost network, and is used to output a 24-hour photovoltaic output prediction curve according to photovoltaic output historical data; The load prediction model is constructed based on a second LSTM network, and is used to output a 24-hour load prediction curve according to load historical data. 3.The energy storage dispatching method based on multi-time scale rolling optimization of claim 1, wherein: In the step S1, the day-ahead prediction model is used to output a 24-hour SOC reference trajectory and a benchmark charging and discharging plan based on a 24-hour electricity price prediction curve, a 24-hour photovoltaic output prediction curve, a 24-hour load prediction curve, and an initial SOC with an hour resolution; the 24-hour SOC reference trajectory serves as a boundary condition of the intra-day prediction model; The objective function output by the day-ahead prediction model is to maximize the value of a whole-cycle benefit function, and is constrained by an energy storage dynamic constraint function, a grid interaction constraint function, a charging and discharging mutual exclusion constraint function, and a photovoltaic output uncertainty robust optimization constraint function; The formula of the whole-cycle benefit function is: Max Profit = Max{λP DA (t)dt} (1) DA (t) represents the price prediction value of the tth hour of the day-ahead; P dischar (t) represents the discharging power of the energy storage system in the tth hour of the day-ahead; P char (t) represents the charging power of the energy storage system in the tth hour of the day-ahead; Δt DA represents the time resolution of the day-ahead prediction model, i.e. 1 hour; C degradation represents the life loss cost based on the Rainflow counting method; C grid represents the grid interaction penalty term for suppressing power peak-valley fluctuations; The formula of the energy storage dynamic constraint function is: wherein SOC(t) represents the SOC of the energy storage system at the tth hour of the day before; SOC(t-1) represents the SOC of the energy storage system at the (t-1)th hour of the day before; η char represents the charging efficiency of the energy storage system; η dischar represents the discharging efficiency of the energy storage system; E rated represents the rated capacity of the energy storage system; The formula of the grid interaction constraint function is: P grid,min (t)≤P grid (t)=P load (t)+P char (t)-P pv_pred (t)-P dischar (t)≤ P grid,max (t); Among them, P grid,min (t) represents the minimum power output of the power grid at hour t before the day; P grid (t) represents the power grid power at hour t before the day; P load (t) represents the load power at hour t before the current day; P pv_pred (t) represents the photovoltaic output forecast value at the tth hour before the day; P grid,max (t) represents the maximum power output of the power grid at hour t before the day; The formula of the charging and discharging mutual exclusion constraint function is: P dischar (t)·P char (t) = 0; The formula of the photovoltaic output uncertainty robust optimization constraint function is: P pv_min (t)≤P pv_pred (t)≤P pv_max (t); P (t) = P (t) - P (t) (1) where P pv_min (t) is the minimum value of the photovoltaic power prediction for the t-th hour of the day ahead; P pv_max (t) is the maximum value of the photovoltaic power prediction for the t-th hour of the day ahead; and P pv_min (t) = 0.9 P pv_pred (t), P pv_max (t) = 1.1 P pv_pred (t) ; The intraday prediction model is used to output a corrected charge-discharge plan based on a 24-hour SOC reference trajectory, a benchmark charge-discharge plan and feedback data with a resolution of 15 minutes; the feedback data includes actual SOC, photovoltaic output deviation and frequency modulation response records; The objective function output by the intraday prediction model is to minimize the value of a prediction deviation function and is constrained by a photovoltaic output constraint function and a power fluctuation rate limiting function; The formula of the prediction deviation function is: Min Cost ID Min Cost represents the minimum value of the prediction deviation of the energy storage system; k is 0, 1, 2, …; λ ID (t) represents the electricity price prediction value of the tth hour in the day; P' dischar (t) represents the discharge power of the energy storage system in the tth hour in the day; α represents a SOC tracking weight coefficient, which is used to control the deviation penalty degree of the 24-hour SOC reference track and the benchmark charge-discharge plan; SOC' ref (t) represents the 24-hour SOC reference track; The formula of the photovoltaic output constraint function is: P pv,ID (t) = P pv,DA (t) + ΔP pv,error (t); where P pv,ID (t) denotes the photovoltaic power value at time t of the day; P pv,DA (t) denotes the photovoltaic prediction value at time t of the day; ΔP pv,error (t) denotes the photovoltaic prediction error value at time t of the day, and ΔP pv,error (t) ~ N(0, σ 2 ), σ denotes the standard deviation of the normal distribution; The formula of the power fluctuation rate limiting function is: |P′ grid (t)-P′ grid (t-1)|≤0.1P′ grid,max ; P′ grid (t) represents the grid power at time t in the day; P′ grid (t-1) represents the grid power at time t-1 in the day; P′ grid,max represents the maximum grid output in the day; The real-time prediction model is used to output a charge-discharge instruction and a frequency modulation output power based on the corrected charge-discharge plan, a dynamic SOC limit value, a real-time electricity price, a real-time photovoltaic output and the actual SOC with a resolution of 1 minute; The objective function output by the real-time prediction model is to maximize the value of an electricity charge income function and is constrained by a real-time power balance function and a frequency modulation power limiting function; The formula of the electricity charge income function is: Max Profitt RT represents the maximum value of electricity revenue; λ RT (t) represents the real-time electricity price; P dischar (t) represents the discharging power of the energy storage system at time t; P char (t) represents the charging power of the energy storage system at time t; Δt RT represents the control period of the real-time prediction model; β represents the frequency modulation revenue weight coefficient; P FR (t) represents the power at time t in response to the frequency modulation demand of the power grid; The formula of the real-time power balance function is: P grid (t) = P load (t) + P char (t) - P pv,actual (t) - P dischar (t) + P FR (t); where P grid (t) denotes the grid power at time t; P load (t) denotes the load power at time t; P char (t) denotes the charging power of the energy storage system at time t; P pv,actual (t) denotes the actual photovoltaic power at time t; The formula of the frequency modulation power limiting function is: 0 < P FR (t) < 0.3P rated ; where P rated represents the rated power of the energy storage system. 4.The energy storage dispatching method based on multi-time scale rolling optimization of claim 1, wherein: The step S6 specifically includes: The output of the day-ahead prediction model is optimized at a frequency of once every 24 hours based on a 24-hour sliding window; the output of the intraday prediction model is optimized at a frequency of once every 15 minutes based on a 4-hour sliding window; and the output of the real-time prediction model is optimized at a frequency of once every 1 minute based on a 15-minute sliding window.

5. A light storage energy regulation system based on multi-time scale rolling optimization, characterized in that: The method comprises the following modules: A model creation module is configured to create and train an electricity price prediction model, a photovoltaic output prediction model, a load prediction model, a day-ahead prediction model, an intraday prediction model and a real-time prediction model; A data acquisition module is configured to acquire historical electricity price data, historical photovoltaic output data, historical load data and an initial SOC of an energy storage system, input the historical electricity price data into the electricity price prediction model to obtain a 24-hour electricity price prediction curve, input the historical photovoltaic output data into the photovoltaic output prediction model to obtain a 24-hour photovoltaic output prediction curve, and input the historical load data into the load prediction model to obtain a 24-hour load prediction curve; A day-ahead prediction module is configured to input the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and the initial SOC into the day-ahead prediction model to obtain a 24-hour SOC reference trajectory and a benchmark charge-discharge plan; An intraday prediction module is configured to acquire feedback data including actual SOC, photovoltaic output deviation and frequency modulation response records, and input the 24-hour SOC reference trajectory, the benchmark charge-discharge plan and the feedback data into the intraday prediction model to obtain a corrected charge-discharge plan; The real-time prediction module is configured to set a SOC dynamic limit value, collect a real-time electricity price and a real-time photovoltaic output, input the corrected charging and discharging plan, the SOC dynamic limit value, the real-time electricity price, the real-time photovoltaic output and an actual SOC into a real-time prediction model, and obtain a charging and discharging instruction and a frequency modulation output power, so that energy storage is regulated and controlled based on the charging and discharging instruction and the frequency modulation output power; The rolling optimization module is configured to rollingly optimize outputs of the day-ahead prediction model, the intra-day prediction model and the real-time prediction model; The deviation compensation module is configured to set a deviation compensation mechanism and compensate the frequency modulation output power based on the deviation compensation mechanism; and the deviation compensation mechanism is specifically: When the real-time photovoltaic output deviation is greater than 15% or the real-time electricity price fluctuation rate is greater than 20%, the charging power and discharging power of the energy storage system are adjusted based on preset priorities; based on the frequency modulation signal issued by the power grid, the energy storage system is switched to a frequency modulation mode, and the energy storage system is controlled to operate in the frequency modulation mode according to P FR (t) = K·Δf(t) response frequency deviation; wherein K represents a frequency modulation coefficient, and Δf(t) represents a power grid frequency offset.

6. The multi-time scale rolling optimization based optical storage energy regulation system of claim 5, wherein: In the model creation module, the electricity price prediction model is constructed based on a first LSTM network and is configured to output a 24-hour electricity price prediction curve according to historical electricity price data; The photovoltaic output prediction model is constructed based on an NWP meteorological network and an XGBoost network and is configured to output a 24-hour photovoltaic output prediction curve according to historical photovoltaic output data; The load prediction model is constructed based on a second LSTM network and is configured to output a 24-hour load prediction curve according to historical load data.

7. The multi-time scale rolling optimization based optical storage energy regulation system of claim 5, wherein: In the model creation module, the day-ahead prediction model is configured to output a 24-hour SOC reference trajectory and a benchmark charging and discharging plan based on the 24-hour electricity price prediction curve, the 24-hour photovoltaic output prediction curve, the 24-hour load prediction curve and an initial SOC with an hour as a resolution; and the 24-hour SOC reference trajectory serves as a boundary condition of the intra-day prediction model; The objective function output by the day-ahead prediction model is to maximize a value of a whole-cycle revenue function and is constrained by an energy storage dynamic constraint function, a power grid interaction constraint function, a charging and discharging mutual exclusion constraint function and a photovoltaic output uncertainty robust optimization constraint function; The formula of the whole-cycle revenue function is: Max Profit = Max{λP DA represents the maximum value of the full-cycle benefit of the energy storage system; λ DA (t) represents the price prediction value of the tth hour of the day-ahead; P dischar (t) represents the discharge power of the energy storage system in the tth hour of the day-ahead; P char (t) represents the charging power of the energy storage system in the tth hour of the day-ahead; Δt DA represents the time resolution of the day-ahead prediction model, i.e. 1 hour; C degradation represents the life loss cost based on the Rainflow counting method; C grid represents the grid interaction penalty term for suppressing power peak-valley fluctuations; The formula of the energy storage dynamic constraint function is: wherein SOC(t) represents the SOC of the energy storage system at the tth hour of the day before; SOC(t-1) represents the SOC of the energy storage system at the (t-1)th hour of the day before; η char represents the charging efficiency of the energy storage system; represents the discharging efficiency of the energy storage system; E rated represents the rated capacity of the energy storage system; The formula of the power grid interaction constraint function is: P grid,min (t)≤P grid (t)=P load (t)+P char (t)-P pv_pred (t)-P dischar (t)≤ P grid,max (t); where P grid,min (t) denotes the minimum grid power at hour t of the day; P grid (t) denotes the grid power at hour t of the day; P load (t) denotes the load power at hour t of the day; P pv_pred (t) denotes the photovoltaic power production forecast at hour t of the day; P grid,max (t) denotes the maximum grid power output at hour t of the day; The formula of the charging and discharging mutual exclusion constraint function is: P dischar (t)·P char (t) = 0; The formula of the photovoltaic output uncertainty robust optimization constraint function is: P pv_min (t)≤P pv_pred (t)≤P pv_max (t); P (t) = P (t) - P (t) (1) where P pv_min (t) is the minimum value of the photovoltaic power prediction for the t-th hour of the day ahead; P pv_max (t) is the maximum value of the photovoltaic power prediction for the t-th hour of the day ahead; and P pv_min (t) = 0.9 P pv_pred (t), P pv_max (t) = 1.1 P pv_pred (t) ; The intra-day prediction model is configured to output a corrected charging and discharging plan based on the 24-hour SOC reference trajectory, the benchmark charging and discharging plan and feedback data with 15 minutes as a resolution; and the feedback data includes an actual SOC, a photovoltaic output deviation and a frequency modulation response record; The objective function output by the intra-day prediction model is to minimize a value of a prediction deviation function and is constrained by a photovoltaic output constraint function and a power fluctuation rate limiting function; The formula of the prediction deviation function is: Min Cost = Min (Min Cost (k) ), k = 0, 1, 2, …, K ID Min Cost (k) = Min (Min Cost (k) ), k = 0, 1, 2, …, K ID (t) represents the power price prediction value of the tth hour in the day; P (t) represents the discharge power of the energy storage system in the tth hour in the day; a represents a SOC tracking weight coefficient, used to control the deviation penalty degree of the 24-hour SOC reference track and the benchmark charge-discharge plan; SOC'(t) represents the SOC of the energy storage system in the tth hour in the day; SOC'(t) represents the SOC of the energy storage system in the tth hour in the day; and ref (t) represents the 24-hour SOC reference track; The formula of the photovoltaic output constraint function is: P pv,ID (t) = P pv,DA (t) + ΔP pv,error (t); P (t) = P (t) + ΔP (t) (1) pv,ID (t) denotes the photovoltaic output value at time t within the day; P pv,DA (t) denotes the photovoltaic prediction value at time t of the day; ΔP pv,error (t) denotes the photovoltaic prediction error value at time t of the day, and ΔP pv,error (t) ~ N(0, σ 2 ), σ denotes the standard deviation of the normal distribution; The formula of the power fluctuation rate limiting function is: |P′ grid (t)-P′ grid (t-1)|≤0.1P′ grid,max ; P′ grid (t) represents the grid power at time t in the day; P′ grid (t-1) represents the grid power at time t-1 in the day; P′ grid,max represents the maximum grid output in the day; The real-time prediction model is configured to output a charging and discharging instruction and a frequency modulation output power based on the corrected charging and discharging plan, a SOC dynamic limit value, a real-time electricity price, a real-time photovoltaic output and an actual SOC with 1 minute as a resolution; The objective function output by the real-time prediction model is to maximize a value of an electricity charge revenue function and is constrained by a real-time power balance function and a frequency modulation power limiting function; The formula of the electricity charge revenue function is: Max Profitt RT represents the maximum value of electricity revenue; λ RT (t) represents the real-time electricity price; P dischar (t) represents the discharging power of the energy storage system at time t; P char (t) represents the charging power of the energy storage system at time t; Δt RT represents the control period of the real-time prediction model; β represents the frequency regulation revenue weight coefficient; P FR (t) represents the power at time t in response to the frequency regulation demand of the power grid; The formula of the real-time power balance function is: P grid (t) = P load (t) + P char (t) - P pv,actual (t) - P dischar (t) + P FR (t); where P grid (t) denotes the grid power at time t; P load (t) denotes the load power at time t; P char (t) denotes the charging power of the energy storage system at time t; P pv,actual (t) denotes the actual photovoltaic power at time t; The formula of the frequency modulation power limit function is: 0 < P FR (t) < 0.3P rated ; where P rated represents the rated power of the energy storage system. 8.The optical storage energy regulation system based on multi-time scale rolling optimization of claim 5, wherein: The rolling optimization module is specifically configured to: The output of the day-ahead prediction model is rolling optimized at a frequency of once every 24 hours based on a 24-hour sliding window; the output of the intra-day prediction model is rolling optimized at a frequency of once every 15 minutes based on a 4-hour sliding window; and the output of the real-time prediction model is rolling optimized at a frequency of once every 1 minute based on a 15-minute sliding window.

Images