Independent energy storage model construction method based on price difference multi-section quotation strategy

Abstract

The invention discloses an independent energy storage model construction method based on a price difference multi-section quotation strategy, and the method comprises the steps: building a prediction model based on machine learning based on a historical whole network load, day-ahead electricity price data, and an operation day whole network load prediction value; the bidding day predicts the electricity price of the operation day (the operation day represents a natural day in which the current electric energy market transaction plan is actually executed, and the bidding day is the day before the operation day). Related basic configuration information (including rated capacity, rated power and the like), running day-ahead electricity price, running day real-time electricity price and system electricity price of independent energy storage are known, independent energy storage running day-ahead total revenue maximization is used as a target function, a plurality of constraint equations such as a charging and discharging physical process and multi-section quotation business constraint are added, a model is solved, and the independent energy storage running day-ahead total revenue maximization model is obtained. And an optimal multi-section quotation strategy based on the day-ahead price difference can be obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of independent energy storage power market, and specifically relates to a method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy. Background Art

[0002] In recent years, with the rapid development of renewable energy power generation, the scale and scope of electricity spot trading have also been expanding. In order to improve the stability and reliability of the power system and meet the needs of environmental protection and energy conservation, independent energy storage has gradually become an important part of participating in electricity spot trading. Since independent energy storage has the advantages of fast response speed, high regulation rate, accurate regulation accuracy, and flexible charge and discharge conversion, participating in market transactions can give full play to its supporting role in ensuring the safe supply of electricity and serving the construction of new power systems. With the participation of independent energy storage in the spot market, how to obtain better returns is a concern for every company participating in this market.

[0003] With the rapid development of science and technology, Artificial Intelligence (AI) is one of the hottest topics in the world. It is the vane leading the development of the world's future science and technology and the transformation of lifestyles in the 21st century, and has been widely used in many fields. Therefore, in this system, operations optimization, machine learning and deep learning technologies are used to build models based on the full amount of data related to market transactions, and automatically generate transaction declaration strategies to achieve the goal of reducing losses and gaining profits.

[0004] At present, the day-ahead declaration method for multi-stage quotations of independent energy storage has only been gradually liberalized. There is no guiding theory on how independent energy storage stations can reasonably complete multi-stage quotations. Most independent energy storage stations only make multi-stage quotations based on the price difference between the day-ahead price and the real-time price. This method is simple and easy to operate, but it does not consider the influence of factors such as the day-ahead charging and discharging period and SOC size.

[0005] To this end, we propose a method for constructing an independent energy storage model based on a price difference multi-segment quotation strategy to solve the above problems. Summary of the invention

[0006] The technical problem to be solved by the present invention is: to provide a method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy, aiming to solve the problem that the day-ahead declaration method for independent energy storage multi-stage quotation has only been gradually liberalized, and there is no guiding theory on how independent energy storage stations can reasonably complete multi-stage quotation. Most independent energy storage stations only make multi-stage quotations based on the price difference between the day-ahead price and the real-time price. This method is simple and easy to operate, but does not consider the impact of factors such as the day-ahead charging and discharging period and the SOC size.

[0007] The objective of the present invention is achieved through the following technical solutions:

[0008] A method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy, specifically comprising the following steps:

[0009] S1. Train the model using the GBDT algorithm;

[0010] S2. First, input the relevant historical data information of the entire network into the 3sigma (3σ) criterion algorithm and the least squares method to divide the data into three sections: low system load rate, average system load rate, and high system load rate. Then, input the system load rate data and related characteristic data into the GBDT model for real-time price prediction. By using the historical data information and weather data of the entire network of independent energy storage, the typical curves of real-time price and system load rate are fitted according to the 3sigma (3σ) criterion algorithm and the least squares method. Then, the box plot algorithm is used to calculate the two segmentation points of the typical curve of real-time price and the two corresponding segmentation points of the typical curve of system load rate corresponding to the segmentation points, thereby dividing the 96 moments of the whole day into three sections: low system load rate, average system load rate, and high system load rate. Then, combined with the data characteristics of each section, data with the same distribution characteristics are obtained from the historical data;

[0011] S3, build XGBoost integrated algorithm;

[0012] S4, inputting the relevant data of the day-ahead electricity price in step S3, and predicting the day-ahead price;

[0013] S5. Input basic configuration parameters related to independent energy storage and basic parameters of the model;

[0014] S6. Establish the objective function and constraints to maximize the benefits;

[0015] S7, solve the model and related constraints in step 6 by using cp l ex in python, and obtain multiple quotation results under this objective function and constraint conditions;

[0016] S8. Output the final multi-stage quotation strategy and the day-ahead electricity declared, which will be used for the independent energy storage multi-stage quotation declaration on D day.

[0017] Furthermore, in step S1, the GBDT algorithm determines the segment to which the moment belongs based on the system load rate at each moment in the future window period, and selects the corresponding training data set from the historical data based on the data characteristics of the segment. Then, multiple decision trees with appropriate depths are established using these data, and each tree is used as a weak learner. The GBDT algorithm gradually generates these weak learners and assigns corresponding weights to each weak learner.

[0018] Furthermore, in step S2, the historical data information of the entire network includes the electricity price data of the entire network, the load data of the entire network, the power transmission data of the entire network, the installed capacity of the entire network and the meteorological data of the relevant dates.

[0019] Furthermore, in step S3, constructing the XGBoost integrated algorithm is specifically as follows: according to the calculated system load rate segmentation point, determine the segment to which the system load rate at each moment in the future window period belongs; then, based on the data characteristics of the segment, select a training data set with a similar segmentation trend from the historical data; then, use these data to train multiple weak learners (decision trees) based on the CART (Classification And Regression Tree) algorithm; each weak learner focuses on the part of the previous model's incorrect prediction in each round of iteration; the XGBoost algorithm corrects the error in the previous round of iteration by gradually constructing a new tree, and assigns corresponding weights to each weak learner; finally, weighted synthesis is performed on the outputs of all weak learners to generate a final strong learner model.

[0020] Furthermore, in step S4, the relevant data of the day-ahead electricity price include market disclosure data, network-wide electricity price data, network-wide load data, network-wide load rate data and meteorological data.

[0021] Furthermore, in step S5, the basic configuration parameters related to the independent energy storage and the basic parameters of the model include rated power, rated capacity, charge-discharge conversion coefficient, peak-valley coefficient, multi-stage quoted output, and maximum and minimum price parameters.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention provides a price difference-based multi-segment quotation optimization strategy based on the predicted day-ahead electricity price and real-time electricity price. In combination with market disclosure data, price forecast data, etc., an integrated learning model is established to predict the day-ahead electricity price and the real-time electricity price. On this basis, the total revenue maximization is taken as the goal, the physical process constraints of charging and discharging and the quotation segment constraints are considered, and the CPLEX solver is used to optimize the multi-segment quotation results of independent energy storage, so as to assist independent energy storage sites in completing the multi-segment quotations submitted on the day-ahead. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is the overall flow chart of the present invention;

[0025] Figure 2 It is a multi-segment quotation diagram in the present invention;

[0026] Figure 3 A diagram of a multi-stage quotation process based on price difference in the present invention;

[0027] Figure 4 A comparison chart of the day-ahead price forecast data and the real-time price forecast data in the present invention;

[0028] Figure 5 This is a curve diagram of the previous day's electricity quantity results in the present invention. DETAILED DESCRIPTION

[0029] The following will describe the implementation methods of the present application in detail with the help of accompanying drawings and examples, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0030] The invention provides a method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy, and its overall structure diagram is as follows: Figure 1 As shown, the following steps are included:

[0031] Step 1: Input relevant historical data information of the entire network: including the electricity price data of the entire network, the load data of the entire network, the power transmission data of the entire network, the installed capacity of the entire network and the meteorological data of the relevant date. The data is divided into three sections: low system load rate, flat system load rate and high system load rate by first inputting the input data into the 3sigma (3σ) criterion algorithm and the least squares method. Then the system load rate data and related characteristic data are input into the GBDT model for real-time price prediction. The detailed construction and prediction process of the model is Step 2.

[0032] Step 2: Use the real-time electricity price and GBDT integrated algorithm to predict the real-time price: By using the historical data information and weather data of the entire network of independent energy storage, the typical curves of the real-time price and the system load rate are fitted according to the 3sigma (3σ) criterion algorithm and the least squares method, and then the box plot algorithm is used to calculate the two segmentation points of the typical curve of the real-time price and the two corresponding segmentation points of the typical curve of the system load rate corresponding to the segmentation point, thereby dividing the 96 moments of the whole day into three segments of low system load rate, average system load rate, and high system load rate. Then, combined with the data characteristics of each segment, data with the same distribution characteristics are obtained from the historical data; the corresponding prediction model is trained by the GBDT algorithm. The GBDT algorithm determines the segment to which the moment belongs based on the system load rate at each moment in the future window period, and selects the corresponding training data set from the historical data based on the data characteristics of the segment. Then, multiple decision trees with appropriate depth are established using these data, and each tree is used as a weak learner. The GBDT algorithm gradually generates these weak learners and assigns corresponding weights to each weak learner. Finally, by applying this model and the segmented correction algorithm of the system load rate, the real-time price within a certain window period in the future can be accurately predicted.

[0033] Step 3: Input the relevant data of the day-ahead electricity price: including market disclosure data, network-wide electricity price data, network-wide load data, network-wide load rate data and meteorological data. The day-ahead price is predicted by inputting the input day-ahead electricity price-related data into the XGBoost integrated algorithm model. The detailed construction and prediction process of the model is Step 4.

[0034] Step 4: Use the XGBoost ensemble algorithm to predict the day-ahead price: First, determine the segment to which the system load rate at each moment in the future window belongs based on the calculated system load rate split point. Next, based on the data features of the segment, select a training data set with a similar segmentation trend from the historical data. Then, use these data to train multiple weak learners (decision trees) based on the CART (Classification And Regression Tree) algorithm. Each weak learner focuses on the part that was incorrectly predicted by the previous model in each round of iteration. The XGBoost algorithm corrects the errors in the previous round of iterations by gradually building new trees and assigning corresponding weights to each weak learner. Finally, the outputs of all weak learners are weighted and integrated to generate the final strong learner model.

[0035] Step 5: Input the basic configuration parameters related to independent energy storage and the basic parameters of the model: rated power, rated capacity, charge-discharge conversion coefficient, peak-valley coefficient, multi-stage quoted output, maximum and minimum prices and other parameters.

[0036] Step 6: Establish an objective function to maximize benefits and constraints: the state before and after SOC, the relationship between the day-ahead declaration and the execution direction, and the charge and discharge power constraints; the upper and lower limits of the bid price, the upper and lower limits of the output value, and the relationship between two adjacent sections of the bid output.

[0037] Step 7: Solving the model: Solve the model and related constraints in Step 6 through cp l ex in Python to obtain multiple quotation results under this objective function and constraints.

[0038] Step 8: Output the final multi-stage quotation strategy and the day-ahead electricity reported, which will be used for the independent energy storage multi-stage quotation report on D day.

[0039] Based on the total revenue of the independent energy storage day-ahead declaration market, the total revenue mainly includes settlement income, electric energy loss costs, and excess profit recovery. The present invention takes the maximization of the total revenue of independent energy storage with real-time and day-ahead market deviation settlement as the objective function, combines the independent energy storage settlement income, day-ahead price forecast data, and real-time price forecast data, and assists users to complete the day-ahead declaration results of multiple quotations to maximize revenue.

[0040] The present invention mainly generates a multi-stage quotation strategy for D-day from 11:00 to 12:00 on D-1. The multi-stage quotation strategy model requires the day-ahead price, real-time price data and other related parameters. Table 1 shows the corresponding input data and output data of the model and their explanations.

[0041] Table 1 Input and output parameter information

[0042]

[0043]

[0044] The objective function is as follows:

[0045] J=J_income t -J_energy_loss t -J_chaoe t (6)

[0047] in:

[0048] J_income t :Settlement income

[0049] J_energy_loss: Electric energy loss cost

[0050] J_chaoe: Excess profit recovery loss fee

[0051] (1) Settlement income

[0052] The settlement income includes two parts: the real-time electricity price on D-day and the electricity charge for the actual electricity consumption, and the day-ahead electricity consumption on D-day and the electricity charge for the day-ahead price minus the real-time price deviation.

[0053] J_income t =(P_riqian t -P_shishi t )*Q_riqian t +P_shishi t *Q_shish t i

[0054] J_income t :Settlement income

[0055] P_riqian t : Today's price forecast

[0056] P_shishi t : Real-time price prediction

[0057] Q_riqian t: Independent energy storage reported power consumption

[0058] Q_shishi t : Independent energy storage simulation execution power

[0059] (2) Electricity loss costs

[0060] When real-time charging is performed on D-day, electric energy loss costs will be incurred. The costs are calculated through real-time electricity consumption, real-time electricity price, peak-valley coefficient and transmission and distribution price, capacity compensation electricity price, line loss cost of the online link, system operation cost and charging and discharging conversion efficiency. The corresponding formula is as follows (charging power is negative, discharging power is positive, so there is a negative sign when calculating electric energy loss in the formula).

[0061] if Q_shishi t <0:

[0062] J_energy_loss t =-Q_shishi t *((transmission_price+capacity_price+

[0063] net_loss_cost+system_run_cost)*peak_valley_rate+P_shishi t )(1-e s_eta)transmission_price: transmission and distribution price

[0064] capacity_price: capacity compensation electricity price

[0065] net_loss_cost: the cost of line loss in the online link

[0066] system_run_cost: system operation electricity cost

[0067] peak_valley_rate: peak-valley coefficient

[0068] es_eta: charge-discharge conversion coefficient

[0069] (3) Excess profit recovery loss

[0070] When real-time charging is performed on D-day, and the day-ahead power consumption is greater than 1.2 times the real-time power consumption or the day-ahead power consumption is less than 0.8 times the real-time power consumption, an excess profit recovery loss fee will be incurred. The fee is calculated as the product of the excess power consumption and the price difference between the day-ahead price and the real-time price. The corresponding formula is as follows:

[0071] if Q_shishi t <0:

[0072] J_chaoe t =-(max(0,P_shishi t -P_riqian t )*min(0,Q_riqian t -1.2*Q_shishi t )

[0073] +max(0,P_riqian t -P_shishi t )*min(0,0.8*Q_shishi t -Q_riqian t ))

[0074] Constraints

[0075] (1) SOC state constraints before and after, that is, the available capacity at the previous moment is the sum of the available capacity at the current moment and the power consumed at the current moment (charging is negative and discharging is positive)

[0076] S_r t =S_rq t-1 -Q_riqian t

[0077] S_shi t =S_shi t-1 -Q_shishi t

[0078] S_r t : Capacity before time t

[0079] S_shi t : Actual capacity at time t

[0080] (2) The direction reported and executed on the previous day must be consistent, that is, the charging and discharging direction of the electricity reported on the previous day must be consistent with the actual charging and discharging direction executed. For example, if the electricity reported on the previous day is charging, the real-time electricity is either charging or neither charging nor discharging, and cannot be discharging.

[0081] 0≤Q_riqian t *Q_shishi t

[0082] (3) At time t, the charge and discharge power cannot exceed its power

[0083] -P*0.25≤Q_riqian t ≤P*0.25

[0084] -P*0.25≤Q_shishit ≤P*0.25

[0085] P: Rated power of independent energy storage

[0086] (4) The price constraints for the two adjacent price segments of the bid submission are that the price for the first segment must be 0.1 (yuan / MWh) higher than the price for the second segment.

[0087] P_list n+1 -P_list n ≥0.1(1≤n≤N-1)

[0088] P_list n Indicates the nth quotation, N indicates the number of charging and discharging output stages

[0089] (5) Constraints on upper and lower limits of bid prices

[0090] P_min≤P_list n ≤P_max(1≤n≤N-1)

[0091] P_max represents the maximum value of the quotation, P_min represents the minimum value of the quotation, and N represents the number of charging and discharging output stages.

[0092] (6) The upper and lower limits of the two adjacent output values ​​reported are constrained, and the previous output value is 2MW greater than the next output value

[0093] Q_list n+1 -Q_list n ≥2(1≤n≤N)

[0094] Q_list n Indicates the nth output value, N indicates the number of charge and discharge output stages

[0095] (7) Upper and lower limits of declared output

[0096] -P≤Q_list n ≤P(1≤n≤N+1)

[0097] Q_list n Indicates the nth output value, P indicates the rated power, and N indicates the number of charge and discharge output stages

[0098] (8) The constraint of reporting the highest charging output value (lowest discharging output value) is that the last charging output value and the first discharging output value are 0.

[0099] Q_list N1+1 =0

[0100] Q_list N1+1 Indicates the N1+1th output value, N1 indicates the number of charging output stages

[0101] (9) Declare the minimum charging output constraint. The first charging output value is the opposite of the maximum rated power.

[0102] Q_list 1 =-P

[0103] Q_list 1 Indicates the first output value

[0104] (10) Declare the maximum discharge output constraint. The final output value of the discharge output is the maximum rated power.

[0105] Q_list N+1 =P

[0106] Q_list N+1 Indicates the N+1th output value

[0107] Where N1 is the number of charging output stages (1≤N1≤5), N is the total number of charging and discharging output stages 2≤N1≤10; P_list is the quoted price corresponding to the declared output, Q_list is the output value of the declared output; P is the maximum power value of the output; P_max is the maximum value of the declared price (the current highest price of Shandong spot electricity price is 1300), and P_min is the minimum value of the declared price (the current lowest price of Shandong spot electricity price is -80);

[0108] (11) Output constraints for bidding

[0109] The charging and discharging output values ​​and quotation forms of multi-stage quotations are as follows: Figure 2 As shown in the figure, the charging and discharging process of the multi-stage quotation is converted into a multi-stage quotation process based on price difference. The multi-stage quotation winning process based on price difference is as follows: Figure 3 As shown, the red curve is the price difference curve, and the stepped shape is the multi-segment quotation of the price difference (symmetrical for charging and discharging). The winning rule is: that is, find the price difference of the day-ahead price at each moment minus the lowest day-ahead price in the 96-point day-ahead forecast price. If there is a winning segment that meets the price difference, the corresponding winning output value is the right endpoint of the output segment.

[0110] According to the multi-segment price difference quotation logic, the output values ​​of charging and discharging are processed as the same direction, that is, the output of the charging side is processed as a positive number. The rules are as follows:

[0111] lb_sd n =-Q_list n ,(1≤n≤N1)

[0112] ub_sd n =-Q_list n+1 ,(1≤n≤N1)

[0113] lb_fdm =Q_list N1+m ,(1≤m≤N-N1)

[0114] ub_fd m =Q_list N1+m+1 ,(1≤m≤N-N1)

[0115] By processing the quoted segment into the output difference logic of the charging and discharging output segment

[0116]

[0117] Declared price value at discharge

[0118] P_list_fd m =P_list N1+m ,(1≤m≤N-N1)

[0119] Declared price value when charging

[0120] P_list_sd n =P_list n ,(1≤n≤N1)

[0121] The difference in declared prices for the same output stages of charging and discharging

[0122] dP m,n =P_list_fd m -P_list_sd n

[0123] The bidding logic for charging with the same price difference and output stages is as follows:

[0124]

[0125] The logic of bidding for the price difference and the same number of output stages is as follows:

[0126]

[0127] Where P_riqian_max t =P_riqian t -min(P_riqian)

[0128] Whether the current price is charged or discharged, logic flag action t (action t ∈{0,1}), action t When it is 1, it means that discharge is possible at the moment t)

[0129] The final charge and discharge output power is Pess t :

[0130] Pess t =action t ×Pess_discharge t +(1-action t )×Pess_charge t .

[0131] Case Analysis

[0132] The present invention establishes a deep learning model for day-ahead and real-time electricity prices based on past characteristic data of Shandong Province. Taking the multi-segment quotations on July 19 as an example, the established day-ahead electricity price machine learning model is first called to predict the day-ahead electricity price and real-time electricity price on July 19. The predicted day-ahead and real-time electricity price data are shown in Table 2 (Table 2 only shows the results for some moments of the day). In order to intuitively show the price difference direction between the day-ahead electricity price and the real-time electricity price, the following plot is drawn: Figure 4 Comparison chart of day-ahead and real-time electricity prices.

[0133] Table 2: Day-ahead price and real-time price prediction results

[0134]

[0135]

[0136] Next, based on the cost parameter information of an independent energy storage station in Shandong Province on July 19, as well as the day-ahead predicted electricity price and real-time predicted electricity price and other related data, the constructed CP-EX was brought in to solve the declaration results of the multi-segment quotation. Table 3 shows the multi-segment quotation output value obtained by the model solution, and Table 4 shows the multi-segment quotation price difference result under the multi-segment quotation output. It can be seen from the table that 8 quotation segments correspond to 8 electricity prices, and the segmented quotation on the charging and discharging side meets the monotonic non-decreasing rule.

[0137] Table 3 Multi-stage quotation output results

[0138]

[0139]

[0140] Table 4. Results of multi-segment quotations based on price spreads

[0141]

[0142] From the output results and price difference relationship data of the above multi-segment quotations, it can be seen that the minimum price difference for clearing the day before the declaration is 157 yuan / MWh, corresponding to an output of 15MW, and the maximum price difference is 457 yuan / MWh, corresponding to an output of 100MW. The maximum price difference between the highest and lowest prices in the day in the day-ahead forecast data is 594.7 yuan / MWh. Therefore, the output value obtained by the model solution is 100MW. The day-ahead electricity is charged at the lowest price of the day before, and discharged at the highest price of the day before. Table 5 shows the cleared electricity at different times under the multi-segment quotation. Figure 5 This is the result curve of the day-ahead electricity quantity. From the result, we can know that the charging and discharging of the day-ahead electricity quantity satisfies the price difference direction between the day-ahead electricity price and the real-time electricity price, and the profit is positive.

[0143] Table 5. Day-ahead power and day-ahead output results at different times

[0144]

[0145]

[0146] The multi-segment quotations and day-ahead electricity obtained by the above model are substituted into the above formula to obtain the corresponding settlement income, excess profit, total income and other results at 24 moments, and the corresponding results are shown in Table 6. From the income results, it can be seen that the settlement income of the day-ahead declaration strategy obtained by the multi-segment quotation model on July 19 was 75144.91 yuan, the total settlement income was 51164.89 yuan, and the income of the day-ahead electricity at 13:00, 21:00, 22:00, and 23:00 was 12095.19 yuan; at 10:00, 11:0, and 12:00, due to real-time charging and the high electricity price at this time, the income was -51882.36 yuan.

[0147] Table 6 Settlement income and total revenue results

[0148]

[0149]

[0150] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the present invention in any form. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention fall within the protection scope of the present invention.

[0151] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are within the scope of the technical solution of the present invention.

Claims

1. A method for constructing an independent energy storage model based on a price difference multi-segment quotation strategy, characterized in that: The steps include: S1. Train the model using the GBDT algorithm; S2. First, input the relevant historical data information of the entire network into the 3sigma (3σ) criterion algorithm and the least squares method to divide the data into three sections: low system load rate, average system load rate, and high system load rate. Then, input the system load rate data and related characteristic data into the GBDT model for real-time price prediction. By using the historical data information and weather data of the entire network of independent energy storage, the typical curves of real-time price and system load rate are fitted according to the 3sigma (3σ) criterion algorithm and the least squares method. Then, the box plot algorithm is used to calculate the two segmentation points of the typical curve of real-time price and the two corresponding segmentation points of the typical curve of system load rate corresponding to the segmentation points, thereby dividing the 96 moments of the whole day into three sections: low system load rate, average system load rate, and high system load rate. Then, combining the data characteristics of each section, data with the same distribution characteristics are obtained from the historical data. S3, build XGBoost integrated algorithm; S4, inputting the relevant data of the day-ahead electricity price in step S3, and predicting the day-ahead price; S5. Input basic configuration parameters related to independent energy storage and basic parameters of the model; S6. Establish the objective function and constraints to maximize the benefits; S7, solving the model and related constraints in step 6 by using cplex in Python, and obtaining multiple quotation results under the objective function and constraints; S8. Output the final multi-stage quotation strategy and the day-ahead electricity declared, which will be used for the independent energy storage multi-stage quotation declaration on D day.

2. The method for constructing an independent energy storage model based on a price difference multi-segment quotation strategy according to claim 1, characterized in that: In step S1, the GBDT algorithm determines the segment to which the moment belongs based on the system load rate at each moment in the future window period, and selects a corresponding training data set from historical data based on the data characteristics of the segment. Then, these data are used to establish multiple decision trees with appropriate depths. Each tree serves as a weak learner. The GBDT algorithm gradually generates these weak learners and assigns corresponding weights to each weak learner.

3. The method for constructing an independent energy storage model based on a price difference multi-segment quotation strategy according to claim 2, characterized in that: In step S2, the historical data information of the entire network includes the electricity price data of the entire network, the load data of the entire network, the power transmission data of the entire network, the installed capacity of the entire network and the meteorological data of the relevant date.

4. The method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy according to claim 1, characterized in that: In the step S3, constructing the XGBoost integrated algorithm is specifically as follows: according to the calculated system load rate segmentation point, determine the segment to which the system load rate at each moment in the future window period belongs, then, based on the data characteristics of the segment, select a training data set with a similar segmentation trend from the historical data, and then use these data to train multiple weak learners (decision trees) based on the CART (Classification And Regression Tree) algorithm, each weak learner focuses on the part of the previous model's incorrect prediction in each round of iteration, and the XGBoost algorithm corrects the error in the previous round of iteration by gradually constructing a new tree, and assigns a corresponding weight to each weak learner, and finally, weighted synthesis of the outputs of all weak learners to generate a final strong learner model.

5. The method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy according to claim 4, characterized in that: In step S4, the relevant data of the day-ahead electricity price include market disclosure data, network-wide electricity price data, network-wide load data, network-wide load rate data and meteorological data.

6. The method for constructing an independent energy storage model based on a price difference multi-stage quotation strategy according to claim 1, characterized in that: In step S5, the basic configuration parameters related to the independent energy storage and the basic parameters of the model include rated power, rated capacity, charge-discharge conversion coefficient, peak-valley coefficient, multi-stage quoted output, and maximum and minimum price parameters.

Images