Multi-time-scale optimization control method for energy storage based on LSTM prediction and correction

Abstract

A multi-time-scale optimization control method for energy storage based on LSTM prediction and correction belongs to the field of energy storage control technology for new energy power grids. It solves the problem of how to reduce the peak-to-valley difference and volatility of the net load curve, alleviate the system peak-shaving pressure, and improve the economic benefits of energy storage. LSTM is used to predict wind power and load data and calculate the predicted net load data. A prediction planning model is constructed based on the predicted net load curve. With the peak-shaving effect as the goal, the charging and discharging power is planned considering the constraints of energy storage capacity and charging and discharging power. Then, based on the difference between the planned charging and discharging power, an objective function that takes into account the net load standard deviation and the energy storage operation benefit is established. The output of the energy storage in the remaining non-operation period is planned under the relevant constraints of the system and energy storage. The valley-filling and peak-shaving power line is corrected online based on the difference between the remaining power in the prediction planning model and the remaining power during the actual operation of the energy storage, so as to optimize the actual output power and SOC of the energy storage and improve the peak-shaving effect and economic benefits.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy grid energy storage control, and relates to an energy storage multi-time scale optimization control method based on LSTM prediction and correction. Background Art

[0002] Amid the global shift to renewable energy, we must accelerate the construction of a new power system to support energy transition and achieve emission reduction goals. Building a new power system dominated by renewable energy is a key pillar for accelerating the development of a clean, low-carbon, safe, and efficient energy system. As the energy transition deepens, the integration of a high proportion of renewable energy will place a significant burden on power system peak shaving and frequency regulation. The daily output characteristics of wind power often contradict load demand, leading to increased peak-to-valley and volatility in load. Energy storage systems can alleviate this system burden.

[0003] Long-short-term memory (LSTM) neural networks are a special type of recurrent neural network (RNN). Compared to RNNs, which often suffer from long-term dependencies, LSTMs incorporate a structure that remembers cell states. Cell states can be modified through output, input, and forget gates to preserve this information over time, improving prediction accuracy when using large training sets. Currently, LSTM predictions have established applications in wind power and load forecasting, as exemplified by the 2018 publication "Optimal Operation of Battery Energy Storage System Considering Distribution System Uncertainty" (Y. Zheng et al., IEEE Trans. Sustain. Energy). However, these predictions still exhibit certain errors, leading to existing research proposing correction methods to address these errors.

[0004] Traditional methods for reducing load peak-valley differences and alleviating system peak-shaving pressure primarily rely on thermal power and gas turbines for peak shaving. These methods place high demands on the peak-shaving capacity of the units, and frequent start-ups and shutdowns can easily lead to resource waste. The 2019 paper "Optimal Energy Storage System Operation for Peak Reduction in a Distribution Network Using a Prediction Interval" (D. Kodaira, W. Jung S. Han, IEEE Transactions on Smart Grid) proposed using energy storage systems (such as lithium batteries) to actively participate in peak shaving, using dynamic programming or optimization algorithms to solve the objective function and plan the charging and discharging of energy storage. However, these studies are mostly based on historical data analysis and lack real-time performance. Therefore, related studies in recent years, such as the 2020 paper "Optimal ESS Scheduling for Peak Shaving of Building Energy Using Accuracy-Enhanced Load Forecast" (J. S. Wang et al., Energies), have proposed combining load forecasting to plan the next-day charging and discharging of energy storage. However, some methods do not consider the impact of forecast errors, while others consider forecast errors but do not fully consider the effectiveness and economic benefits of valley filling and peak shaving. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to design a multi-time-scale optimization control method for energy storage based on LSTM prediction and correction to reduce the peak-to-valley difference and volatility of the net load curve, alleviate the system peak-shaving pressure and improve the economic benefits of energy storage.

[0006] The multi-time-scale optimization control method for energy storage based on LSTM prediction and correction includes the following steps:

[0007] S1. Use the historical data of wind power and load as training sets to perform LSTM prediction on the next day's wind power and load, and calculate the predicted net load based on the predicted wind power and load;

[0008] S2. Under the condition that the system power balance constraint and the energy storage power constraint are met, a prediction planning model is constructed. The construction process is as follows:

[0009] a) Based on the predicted net load, with peak-shaving effects as the goal, and taking into account the capacity and charge / discharge power limits of energy storage, plan valley-filling power lines and peak-shaving power lines, calculate the planned charge and discharge capacities, and determine whether the planned charge and discharge capacities meet the capacity constraints;

[0010] b) Based on the difference between the planned charge and discharge amounts, an objective function is established that takes into account both the net load standard deviation and the energy storage operation benefit, and the power of the energy storage during the remaining non-operation period is calculated;

[0011] S3. Calculate the planned remaining capacity and the actual remaining capacity of the energy storage. Based on the difference between the actual remaining capacity and the planned remaining capacity, calibrate the valley filling and peak shaving power line online to optimize the actual charging and discharging power and SOC of the energy storage.

[0012] The method of the present invention uses a long-term and short-term neural network with a memory cell structure to predict wind power and load data and calculate predicted net load data; a prediction planning model is constructed based on the predicted net load curve, and with the peak-shaving effect as the goal, the charging and discharging time and power are planned in consideration of the limitations of energy storage capacity and charging and discharging power; then, based on the difference between the planned charging and discharging amounts, an objective function that takes into account both the net load standard deviation and the energy storage operation benefit is established; and under the relevant constraints of the system and energy storage, a PSO algorithm is used to plan the output of the energy storage in the remaining non-operation period; and based on the difference between the remaining power in the prediction planning model and the remaining power during the actual operation of the energy storage, the valley filling and peak shaving power line is corrected online, the actual output period, power and SOC of the energy storage are optimized, and the peak-shaving effect and economic benefit are improved.

[0013] Furthermore, the cell state update formula of the LSTM structure predicted by the LSTM in step S1 is as follows:

[0014]

[0015] Among them, n k 、e k 、o k 、 They are the forget gate, input gate, output gate and cell state candidate value, x k 、c k 、m k are the input, cell state and output at the current moment, c k-1 、m k-1 are the cell state and output at the previous moment, a n 、a e 、a c 、a o is the weight vector, d n d e ,、d c is the corresponding bias vector, σ and tan m are activation functions, and ⊙ is the Hadamard product.

[0016] Furthermore, the calculation formula for calculating the predicted net load based on the predicted wind power and load in step S1 is as follows:

[0017] Pnl,pre,t =P l,pre,t -P w,pre,t (2)

[0018] Among them, P nl,pre,t 、P w,pre,t 、P l,pre,t are the values ​​of predicted net load, predicted wind power and predicted load at time t respectively.

[0019] Furthermore, the method for calculating the planned charging amount and the planned discharging amount described in step S2 is specifically as follows:

[0020] The sum of the predicted minimum net load and the rated power of the lithium battery is used as the valley-filling power line to calculate the planned charging capacity of the energy storage at this time. The calculation formula is as follows:

[0021]

[0022] The difference between the predicted maximum net load and the rated power of the lithium battery is used as the peak shaving power line to calculate the planned discharge capacity of the energy storage at this time. The calculation formula is as follows:

[0023]

[0024] Among them, E c,plan and E d,plan are the planned charging amount and the planned discharging amount respectively, t1 and t2 are the charging periods corresponding to the intersection of the valley filling power line and the load curve, t3 and t4 are the charging periods corresponding to the intersection of the peak shaving power line and the load curve, η c ,η c are the charging efficiency and discharging efficiency, P nl,pre,min and P nl,pre,max are the maximum and minimum values ​​of the predicted net load, P b is the rated power of the lithium battery.

[0025] Furthermore, the method described in step S2 for determining whether the planned charge capacity and the planned discharge capacity meet the capacity constraint condition is as follows:

[0026] Determine whether the planned charging capacity meets the first capacity constraint. If not, move the valley-filling power line downward with a step size of ΔP until the constraint is met.

[0027] The first capacity constraint relationship condition is as follows:

[0028] 0 <SOC max -SOC min -E c,plan / E b <ε0 (5)

[0029] Determine whether the planned discharge capacity meets the second capacity constraint condition. If not, move the peak-shaving power line upward with a step size of ΔP until the constraint condition is met.

[0030] The second capacity constraint relationship condition is as follows:

[0031] 0 <SOC max -SOC min -E d,plan / E b <ε0 (6)

[0032] Among them, SOC max and SOC min They are the upper and lower limits of lithium battery SOC, E b is the rated capacity of the lithium battery, and ε0 is a positive value close to 0.

[0033] Furthermore, the method described in step S2 for establishing an objective function that takes into account both the net load standard deviation and the energy storage operation benefit based on the difference between the planned charging amount and the planned discharging amount, and solving the power of the energy storage during the remaining non-operation period is as follows:

[0034] If the planned charging capacity is greater than the planned discharging capacity, the objective function is to maximize the improvement of the discharge benefit and the predicted net load standard deviation. After normalization, the formula is as follows:

[0035]

[0036] Where n is the number of time periods in the remaining unoperated period, I d * and S d * They are the normalized discharge benefit and the improvement of the predicted net load standard deviation, p t is the peak-valley electricity price, p max is the peak electricity price, P d,t is the discharge power of the lithium battery at time t, S nl,pre and P nl,pre,av are the standard deviation and mean of the predicted net load, maxF1 * is the objective function in this case;

[0037] If the planned charging capacity is less than the planned discharging capacity, the objective function is to minimize the standard deviation of charging cost and predicted net load. After normalization, the formula is as follows:

[0038]

[0039] Among them, I c * and S c *are the normalized charging cost and the standard deviation of the predicted net load, p min is the peak electricity price, P c,t is the charging power of the lithium battery at time t, maxF2 * is the objective function in this case.

[0040] Furthermore, the system power balance constraint and energy storage power constraint described in step S2 are specifically as follows:

[0041] System power balance constraints:

[0042] P g =P load -P wind +P bat (9)

[0043] Among them, P g is the tie line power, P wind is the wind power grid-connected power, P load is the load power, P bat The charging and discharging power of lithium batteries;

[0044] Energy storage power constraints:

[0045]

[0046] Among them, SOC t is the SOC of energy storage at time t.

[0047] Furthermore, the formula for calculating the planned remaining power and the actual remaining power of the energy storage in step S3 is as follows:

[0048]

[0049]

[0050] Among them, E res,plan,t and E res,real,t They are the planned remaining capacity of energy storage at time t and the actual remaining capacity of energy storage; P c,plan,t and P d,plan,t are the planned charging power and the planned discharging power of the energy storage at time t; P c,plan,t and P d,plan,t are the actual charging power and actual discharging power of the energy storage at time t respectively.

[0051] Furthermore, the formula for the difference between the actual remaining power and the planned remaining power in step S3 is as follows:

[0052] E s,t =E res,plan,t -E res,real,t(13)

[0053] Among them, E s,t The difference between the actual remaining power and the planned remaining power.

[0054] Furthermore, the method for online correction of the valley filling and peak clipping power line in step S3 is as follows:

[0055] When E s,t >ε, if the planned charging and discharging power of the energy storage is less than or equal to 0, then the planned valley-filling power line will be increased by 5%P b Otherwise, the peak power line will be reduced by 5%P b ;

[0056] When E s,t >-ε, if the planned charging and discharging power of the energy storage is less than or equal to 0, the planned peak-shaving power line at this time will be increased by 5%P b Otherwise, reduce the valley filling power line by 5% P b ;

[0057] When|E s,t When |≤ε, the valley filling and peak shaving power line is not changed;

[0058] Among them, ε is the allowable error range of power, which is 5% of the rated capacity of the lithium battery.

[0059] The advantages of the present invention are:

[0060] The method of the present invention uses a long-term and short-term neural network with a memory cell structure to predict wind power and load data and calculate predicted net load data; a prediction planning model is constructed based on the predicted net load curve, and with the peak-shaving effect as the goal, the charging and discharging time and power are planned in consideration of the limitations of energy storage capacity and charging and discharging power; then, based on the difference between the planned charging and discharging amounts, an objective function that takes into account both the net load standard deviation and the energy storage operation benefit is established; and under the relevant constraints of the system and energy storage, a PSO algorithm is used to plan the output of the energy storage in the remaining non-operation period; and based on the difference between the remaining power in the prediction planning model and the remaining power during the actual operation of the energy storage, the valley filling and peak shaving power line is corrected online, the actual output period, power and SOC of the energy storage are optimized, and the peak-shaving effect and economic benefit are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flowchart of a multi-time-scale optimization control method for energy storage based on LSTM prediction and correction according to the first embodiment of the present invention;

[0062] Figure 2 This is a diagram of the cell structure of a long short-term memory neural network (LSTM) according to the first embodiment of the present invention;

[0063] Figure 3is a flow chart of prediction error correction according to the first embodiment of the present invention;

[0064] Figure 4 This is a comparison diagram of the valley filling and peak shaving effects before and after error correction of the energy storage multi-time scale optimization control method based on LSTM prediction and correction in Example 1 of the present invention;

[0065] Figure 5 This is a comparison diagram of energy storage SOC before and after error correction of the energy storage multi-time scale optimization control method based on LSTM prediction and correction in Example 1 of the present invention;

[0066] Figure 6 This is a comparison diagram of energy storage power before and after error correction of the energy storage multi-time scale optimization control method based on LSTM prediction and correction in Example 1 of the present invention;

[0067] Figure 7 This is a comparison chart of the valley filling and peak shaving effects of the energy storage multi-time scale optimization control method based on LSTM prediction and correction in Example 1 of the present invention, the constant power strategy, and the power difference strategy;

[0068] Figure 8 This is a comparison chart of energy storage power between the energy storage multi-time scale optimization control method based on LSTM prediction and correction and the constant power strategy and the power difference strategy according to the first embodiment of the present invention;

[0069] Figure 9 This is a comparison chart of evaluation indicators of the energy storage multi-time scale optimization control method based on LSTM prediction and correction in Example 1 of the present invention, the constant power strategy, and the power difference strategy. DETAILED DESCRIPTION

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0071] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0072] Example 1

[0073] like Figure 1 As shown in FIG, the multi-time-scale optimization control method for energy storage based on LSTM prediction and correction includes the following steps:

[0074] Step 1: Input a year's worth of wind power and load data for a particular location as a dataset. Select historical data from one month of spring as a training set and input it into the LSTM memory cell structure. Perform LSTM predictions on the next day's wind power and load. Calculate the predicted net load based on the predicted wind power and load data.

[0075] Step 1.1, such as Figure 2 As shown in the figure, after inputting historical data into the LSTM structure, the update formula of its cell state is as follows:

[0076]

[0077] Among them, n k 、e k 、o k 、 They are forget gate, input gate, output gate and cell state candidate value respectively; x k 、c k 、m k are the input, cell state and output at the current moment respectively; c k-1 、m k-1 are the cell state and output at the previous moment respectively; a n 、a e 、a c 、a o is the weight vector, d n d e ,、d c is the corresponding bias vector; σ and tan m are activation functions; ⊙ is the Hadamard product.

[0078] Step 1.2: Calculate the predicted net load based on the predicted wind power and load data. The calculation formula is as follows:

[0079] P nl,pre,t =P l,pre,t -P w,pre,t (2)

[0080] Among them, P nl,pre,t 、P w,pre,t 、P l,pre,t are the values ​​of predicted net load, predicted wind power and predicted load at time t respectively.

[0081] Step 2: Based on the predicted net load obtained in Step 1, with peak-shaving as the goal, consider the energy storage capacity and charge / discharge power limits to plan valley-filling and peak-shaving power lines, calculate the planned charge and discharge times and amounts, and then, based on the difference between the planned charge and discharge amounts, establish an objective function that takes into account both the net load standard deviation and the energy storage operating benefit. The PSO algorithm is used to calculate the energy storage power during the remaining non-operation period, thus constructing a forecasting and planning model. This construction process must satisfy the relevant constraints of the system and the energy storage.

[0082] Step 2.1: Calculate the planned charging and discharging time and power. The calculation process is as follows:

[0083] The sum of the predicted minimum net load and the rated power of the lithium battery is used as the valley filling power line to calculate the planned charging capacity of the energy storage at this time. The formula is shown in (3). The difference between the predicted maximum net load and the rated power of the lithium battery is used as the peak shaving power line to calculate the planned discharge capacity of the energy storage at this time. The formula is shown in (4):

[0084]

[0085]

[0086] Among them, E c,plan and E d,plan are the planned charging amount and the planned discharging amount respectively; t1 and t2 are the charging periods corresponding to the intersection of the valley filling power line and the load curve; t3 and t4 are the charging periods corresponding to the intersection of the peak shaving power line and the load curve; η c ,η c are charging efficiency and discharging efficiency respectively; P nl,pre,min and P nl,pre,max are the maximum and minimum values ​​of the predicted net load respectively; P b is the rated power of the lithium battery.

[0087] Determine whether the planned charging capacity meets the capacity constraint relationship condition shown in formula (5). If not, move the valley-filling power line downward with a step size of ΔP until the constraint condition is met.

[0088] 0 <SOC max -SOC min -E c,plan / E b <ε0 (5)

[0089] Determine whether the planned discharge capacity meets the capacity constraint relationship condition shown in formula (6). If not, move the peak-shaving power line upward with a step size of ΔP until the constraint condition is met.

[0090] 0 <SOC max -SOC min -E d,plan / E b <ε0 (6)

[0091] Among them, SOC max and SOC min They are the upper and lower limits of the lithium battery SOC; E b is the rated capacity of the lithium battery; ε0 is a positive value close to 0.

[0092] Step 2.2: Based on the difference between the planned charge and discharge amounts, establish an objective function that takes into account both the net load standard deviation and the energy storage operation benefit. Plan the energy storage in the remaining non-operation period T. s The process is as follows:

[0093] If the planned charging capacity is greater than the planned discharging capacity, the objective function is to maximize the improvement of the discharge benefit and the predicted net load standard deviation. After normalization, the formula is as follows:

[0094]

[0095] Where n is the number of time periods in the remaining unoperated period; I d * and S d * are the normalized discharge benefit and the improvement of the predicted net load standard deviation; p t is the peak-valley electricity price; p max is the peak electricity price; P d,t is the discharge power of the lithium battery at time t; S nl,pre and P nl,pre,av are the standard deviation and mean of the predicted net load respectively; maxF1 * is the objective function in this case.

[0096] If the planned charging capacity is less than the planned discharging capacity, the objective function is to minimize the standard deviation of charging cost and predicted net load. After normalization, the formula is as follows:

[0097]

[0098] Among them, I c * and S c * are the normalized charging cost and the standard deviation of the predicted net load; p min is the peak electricity price; P c,t is the charging power of the lithium battery at time t; maxF2 * is the objective function in this case.

[0099] The following constraints must be met during steps 2.3, 2.1, and 2.2:

[0100] System power balance constraints:

[0101] P g =P load -P wind +P bat (9)

[0102] Among them, P g is the tie line power, Pwind is the wind power grid-connected power, P load is the load power, P bat It is the charging and discharging power of lithium battery.

[0103] Energy storage power constraints:

[0104]

[0105] Among them, SOC t is the SOC of energy storage at time t.

[0106] Step 3: Figure 3 As shown in FIG, the planned remaining capacity and the actual remaining capacity of the energy storage are calculated according to the data planned in step 2, and an optimization control method for online correction of the valley filling and peak shaving power line is proposed based on the difference between the two, thereby optimizing the actual charging and discharging power and SOC of the energy storage.

[0107] Step 3.1: Calculate the planned remaining energy storage capacity and the actual remaining energy storage capacity based on the data planned in step 2. The calculation formula is as follows:

[0108]

[0109]

[0110] Among them, E res,plan,t and E res,real,t They are the planned remaining capacity of energy storage at time t and the actual remaining capacity of energy storage; P c,plan,t and P d,plan,t are the planned charging power and the planned discharging power of the energy storage at time t; P c,plan,t and P d,plan,t They are the actual charging power and actual discharging power of the energy storage at time t. The actual charging and discharging power of the energy storage can be calculated based on the difference between the planned valley-filling and peak-shaving power line and the actual net load.

[0111] Step 3.2: Based on the difference between the actual remaining power and the planned remaining power, an optimization control method for online correction of the valley filling and peak shaving power line is proposed. The process is as follows:

[0112] Calculate the difference between the actual remaining power and the planned remaining power:

[0113] E s,t =E res,plan,t -E res,real,t (13)

[0114] Online correction of valley filling and peak shaving power line: When E s,t >ε, if the planned charging and discharging power of the energy storage is less than or equal to 0, then the planned valley-filling power line will be increased by 5%P bOtherwise, the peak power line will be reduced by 5%P b . When E s,t >-ε, if the planned charging and discharging power of the energy storage is less than or equal to 0, the planned peak-shaving power line at this time will be increased by 5%P b Otherwise, reduce the valley filling power line by 5% P b . When|E s,t When |≤ε, the valley filling and peak shaving power line is unchanged. ε is the allowable error range of the power, which is 5% of the rated capacity of the lithium battery.

[0115] In this embodiment, the rated capacity of the lithium battery is set to 6MW, the rated power is set to 2MW, the charging efficiency and the discharging efficiency are both 0.9, and the minimum state of charge (SOC min ) is 0.1, the minimum state of charge (SOC max ) is 0.9; the system sampling interval is 15 minutes, the number of samples per day is 96; ΔP is 0.001, and ε0 is 0.01.

[0116] like Figure 4 、 Figure 5 and Figure 6 As shown in the figure, the valley filling and peak shaving effect, SOC, and power comparison of the method of the present invention before and after online correction are shown. It can be seen from the figure that the SOC of the lithium battery before correction will exceed the upper limit of 0.9, and at the same time cannot recover the initial power, and cannot meet the charge and discharge balance of the day, while the SOC after correction remains in the normal working range.

[0117] like Figure 7 and Figure 8 As shown, a comparison diagram of the valley filling and peak shaving effects and energy storage power of the method of the present invention, the constant power strategy and the power difference strategy is shown. It can be seen from the figure that the valley filling and peak shaving effect of the method of the present invention is the best, and the capacity of the lithium battery can be fully utilized for valley filling and peak shaving.

[0118] like Figure 9 As shown in the figure, the evaluation index comparison chart of the strategy proposed in the present invention, the constant power strategy and the power difference strategy shows that from the peak-shaving effect, the actual net load peak-to-valley difference rate is 34.12%, and the strategy proposed in the present invention is 26.01%, which is a decrease of 8.11%, which is the largest reduction among the three strategies. In addition, the standard deviation of the strategy proposed in the present invention is 2.23, which is a decrease of 0.52, while the other two strategies only decreased by 0.21 and 0.38. This shows that this strategy can effectively reduce the volatility of the net load. From an economic point of view, the energy storage operation income of the strategy proposed in the present invention is 9,658.1 yuan, while the constant power strategy and the power difference strategy are only 9,289.8 yuan and 7,678.3 yuan. This shows that the strategy proposed in the present invention has certain advantages in valley filling and peak shaving effects and economic benefits.

[0119] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A multi-time-scale optimization control method for energy storage based on LSTM prediction and correction, characterized by: The following steps are involved: S1. Use the historical data of wind power and load as training sets to perform LSTM prediction on the next day's wind power and load, and calculate the predicted net load based on the predicted wind power and load; The cell state update formula of the LSTM structure of the LSTM prediction is as follows: Among them, n k 、e k 、o k 、 They are the forget gate, input gate, output gate and cell state candidate value, x k 、c k 、m k are the input, cell state and output at the current moment, c k-1 、m k-1 are the cell state and output at the previous moment, a n 、a e 、a c 、a o is the weight vector, d n d e ,、d c is the corresponding bias vector, σ and tan m are activation functions, and ⊙ is the Hadamard product; S2. Under the condition that the system power balance constraint and the energy storage power constraint are met, a prediction planning model is constructed. The construction process is as follows: a) Based on the predicted net load, with peak-shaving effects as the goal, and taking into account the capacity and charge / discharge power limits of energy storage, plan valley-filling power lines and peak-shaving power lines, calculate the planned charge and discharge capacities, and determine whether the planned charge and discharge capacities meet the capacity constraints; The method for calculating the planned charge capacity and the planned discharge capacity is as follows: The sum of the predicted minimum net load and the rated power of the lithium battery is used as the valley-filling power line to calculate the planned charging capacity of the energy storage at this time. The calculation formula is as follows: The difference between the predicted maximum net load and the rated power of the lithium battery is used as the peak shaving power line to calculate the planned discharge capacity of the energy storage at this time. The calculation formula is as follows: Among them, E c,plan and E d,plan are the planned charging amount and the planned discharging amount respectively, t1 and t2 are the charging periods corresponding to the intersection of the valley filling power line and the load curve, t3 and t4 are the charging periods corresponding to the intersection of the peak shaving power line and the load curve, η c ,η c are the charging efficiency and discharging efficiency, P nl,pre,min and P nl,pre,max are the maximum and minimum values ​​of the predicted net load, P b is the rated power of the lithium battery; b) Based on the difference between the planned charge and discharge amounts, an objective function is established that takes into account both the net load standard deviation and the energy storage operation benefit, and the power of the energy storage during the remaining non-operation period is calculated; S3. Calculate the planned remaining capacity and the actual remaining capacity of the energy storage. Based on the difference between the actual remaining capacity and the planned remaining capacity, calibrate the valley filling and peak shaving power line online to optimize the actual charging and discharging power and SOC of the energy storage.

2. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 1 is characterized in that: The calculation formula for calculating the predicted net load based on the predicted wind power and load in step S1 is as follows: P nl,pre,t =P l,pre,t -P w,pre,t (2) Among them, P nl,pre,t 、P w,pre,t 、P l,pre,t are the values ​​of predicted net load, predicted wind power and predicted load at time t respectively.

3. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 2 is characterized in that: The method for determining whether the planned charge capacity and the planned discharge capacity meet the capacity constraint conditions in step S2 is as follows: Determine whether the planned charging capacity meets the first capacity constraint. If not, move the valley-filling power line downward with a step size of ΔP until the constraint is met. The first capacity constraint relationship condition is as follows: 0<SOC max -SOC min -IN c,plan / IN b <ε0 (5) Determine whether the planned discharge capacity meets the second capacity constraint condition. If not, move the peak-shaving power line upward with a step size of ΔP until the constraint condition is met. The second capacity constraint relationship condition is as follows: 0<SOC max -SOC min -IN d,plan / IN b <ε0 (6) Among them, SOC max and SOC min They are the upper and lower limits of lithium battery SOC, E b is the rated capacity of the lithium battery, and ε0 is a positive value close to 0.

4. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 3 is characterized in that: The method described in step S2 for establishing an objective function that takes into account both the net load standard deviation and the energy storage operation benefit based on the difference between the planned charging amount and the planned discharging amount, and solving the power of the energy storage during the remaining non-operation period, is as follows: If the planned charging capacity is greater than the planned discharging capacity, the objective function is to maximize the improvement of the discharge benefit and the predicted net load standard deviation. After normalization, the formula is as follows: Where n is the number of time periods in the remaining unoperated period, I d * and S d * They are the normalized discharge benefit and the improvement of the predicted net load standard deviation, p t is the peak-valley electricity price, p max is the peak electricity price, P d,t is the discharge power of the lithium battery at time t, S nl,pre and P nl,pre,av are the standard deviation and mean of the predicted net load, maxF1 * is the objective function in this case; If the planned charging capacity is less than the planned discharging capacity, the objective function is to minimize the standard deviation of charging cost and predicted net load. After normalization, the formula is as follows: Among them, I c * and S c * are the normalized charging cost and the standard deviation of the predicted net load, p min is the peak electricity price, P c,t is the charging power of the lithium battery at time t, maxF2 * is the objective function in this case.

5. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 4 is characterized in that: The system power balance constraint conditions and energy storage power constraint conditions described in step S2 are specifically as follows: System power balance constraints: P g =P load -P wind +P bat (9) Among them, P g is the tie line power, P wind is the wind power grid-connected power, P load is the load power, P bat The charging and discharging power of lithium batteries; Energy storage power constraints: Among them, SOC t is the SOC of energy storage at time t.

6. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 5 is characterized in that: The formula for calculating the planned remaining energy storage capacity and the actual remaining energy storage capacity in step S3 is as follows: Among them, E res,plan,t and E res,real,t They are the planned remaining capacity of energy storage at time t and the actual remaining capacity of energy storage; P c,plan,t and P d,plan,t are the planned charging power and the planned discharging power of the energy storage at time t; P c,plan,t and P d,plan,t are the actual charging power and actual discharging power of the energy storage at time t respectively.

7. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 6 is characterized in that: The formula for the difference between the actual remaining power and the planned remaining power in step S3 is as follows: AND s,t =And res,plan,t -AND res,real,t (13) Among them, E s,t The difference between the actual remaining power and the planned remaining power.

8. The energy storage multi-time scale optimization control method based on LSTM prediction and correction according to claim 7 is characterized in that: The method for online correction of the valley filling and peak clipping power line in step S3 is as follows: When E s,t >ε, if the planned charging and discharging power of the energy storage is less than or equal to 0, then the planned valley-filling power line will be increased by 5%P b Otherwise, the peak power line will be reduced by 5%P b ; When E s,t >-ε, if the planned charging and discharging power of the energy storage is less than or equal to 0, the planned peak-shaving power line at this time will be increased by 5%P b ; Otherwise, reduce the valley filling power line by 5% P b ; When|E s,t When |≤ε, the valley filling and peak shaving power line is not changed; Among them, ε is the allowable error range of power, which is 5% of the rated capacity of the lithium battery.

Images