Operation control method of electrochemical energy storage power station suitable for spot trading

Abstract

This electrochemical energy storage power station operation control method, suitable for spot trading, includes: collecting and processing raw data; decomposing the data using ensemble empirical mode decomposition (EEMD) and training the decomposed signals using a Gray Wolf-LSMT network; making predictions and corrections; establishing a power balance formula for the power grid based on a generative adversarial network (GAN); building a total objective function model, using a generator to output prediction scenarios for the total objective function, generating and reducing scenarios; and establishing constraints. Real-time predictions are performed using the LSMT and Gray Wolf algorithms, and objective functions are established using a GAN to determine the long-term revenue and grid stability indicators of the electrochemical energy storage station. The optimal control method for the electrochemical energy storage station is determined, ensuring improved charging and discharging stability of the electrochemical energy storage station, optimizing the operating efficiency of the power plant's power units, maximizing economic benefits, achieving a real-time dynamic balance between the electrochemical energy storage station and the grid's supply and demand, and increasing the absorption rate of new energy.

Description

Technical Field

[0001] The present invention relates to the field of energy storage technology, and in particular to an operation control method of an electrochemical energy storage power station suitable for spot trading. Background Art

[0002] Existing electrochemical energy storage technologies, such as lithium-ion battery energy storage systems, show great potential for participating in grid spot trading, shaving peak loads, and accommodating renewable energy. Electrochemical energy storage stores renewable energy and off-peak electricity generation, discharging it into the grid during peak hours to reduce power plant load. This allows power plant units to balance power across time periods and improve renewable energy absorption. However, electrochemical energy storage suffers from significant time-dependent degradation, meaning that its storage capacity and discharge performance gradually decrease with increasing charge and discharge cycles. Furthermore, near the point of exhaustion, electrochemical energy storage devices often fail to provide a stable discharge current, directly impacting the ability of distributed electrochemical energy storage to provide stable power reserves for the grid system. Consequently, the grid struggles to precisely regulate the amount of power provided to the power plant to its optimal operating power, resulting in reduced grid power quality and suboptimal operating efficiency of the power plant's generating units.

[0003] On the other hand, there's a conflicting goal between maximizing the economic benefits of electrochemical energy storage and ensuring the highest possible grid stability and quality. To address this challenge, it's necessary to develop a system and method that can precisely control electrochemical energy storage based on its real-time power characteristics and real-time data from the grid's supply and demand sides. This system and method should be able to form effective point-control strategies and algorithms to ensure that while meeting power demand, the grid's power supply quality is improved and the power plant's power units operate at optimal operating conditions. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an electrochemical energy storage power station operation control method suitable for spot trading, which realizes the charging and discharging adjustment of the electrochemical energy storage power station in the power grid within a single cycle according to the real-time prediction of wind and solar power output and power load, so as to ensure that the long-term benefits of the electrochemical energy storage power station, the power supply stability of the power grid and the operating status of the power plant are in an optimal balance.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] An electrochemical energy storage power station operation control method suitable for spot trading includes the following steps:

[0007] S1. Collection and processing of raw data; raw data includes historical wind and solar power generation, historical power load, peak and valley metering periods, number and characteristic parameters of generator sets, and number and capacity of electrochemical energy storage groups;

[0008] S2, based on the historical power load data and historical wind and solar power generation in S1, uses ensemble empirical mode decomposition to decompose the data, and uses the Grey Wolf-LSMT network to train the decomposed signal;

[0009] S3: Use the model trained in S2 to predict and correct the current power load to obtain the power load prediction value of each component. Wind and photovoltaic power generation , the predicted values ​​of each component signal are superimposed to obtain the real-time power load prediction value Wind and photovoltaic power generation and make corrections;

[0010] S4. Based on the generative adversarial network, the power balance formula of the power grid is established by combining the power load forecast value in S3, the wind and photovoltaic power generation power forecast value in S4, and the charge and discharge status of the electrochemical energy storage;

[0011] S5. Based on the balance formula, the single-cycle economic benefit objective function formula of electrochemical energy storage and the grid quality evaluation formula are established to obtain the long-cycle net benefit objective function. , for the long-term net profit objective function The total objective function model is obtained by setting weights for the grid stability evaluation formula, and the prediction scenarios of the total objective function are output by the generator to generate scenarios and perform scenario reduction.

[0012] S6. Establish constraints for wind and solar power generation and electrochemical energy storage control, and output optimal electrochemical energy storage power station operation control decisions.

[0013] A control method for electrochemical energy storage power plants suitable for spot trading is proposed. This method includes raw data collection and processing, data prediction based on ensemble empirical mode decomposition (EEMD) and a gray wolf-LSTM network, prediction and correction of power load and wind and solar power generation, establishment of a grid power balance formula based on a generative adversarial network (GAN), construction of economic benefit and grid quality evaluation formulas, establishment of an overall objective function model, and output of optimal control decisions. By combining deep learning and optimization algorithms, this method achieves precise control of electrochemical energy storage power plants, aiming to balance grid supply and demand, improve power supply quality, and power plant operational efficiency. Data decomposition and training using the EEMD and gray wolf-LSTM network improve the accuracy of power load and wind and solar power generation predictions. The GAN-based grid power balance formula enables real-time adjustment of the electrochemical energy storage plant's charging and discharging strategies to ensure grid supply and demand balance. By constructing an economic benefit objective function, a grid quality evaluation formula, and a total objective function model, dual optimization of economic benefit and grid quality is achieved.

[0014] In the above S1, the original data is collected by collecting data at intervals, and the processing is carried out by eliminating obvious outliers and filling in missing values.

[0015] The specific steps in the above S2 are:

[0016] S2.1. Processed power load data Add a standard normal distribution of white noise , historical wind and solar power generation capacity Add a standard normal distribution of white noise , the specific formula is as follows:

[0017] ;

[0018] ;

[0019] In the above formula Indicates the The white noise sequence added is Indicates the The white noise sequence added is Indicates the The additional noise signal of the experiment, Indicates the Additional noise signal for the experiment;

[0020] S2.2. New signal sequences Perform empirical mode decomposition EMD and get modal components and 1 residual component:

[0021] ;

[0022] in, for The first transient components, is the residual component, Represents the average trend of the power load signal;

[0023] For new signal sequences Perform empirical mode decomposition EMD and get modal components and 1 residual component:

[0024] ;

[0025] in, for The first transient components, is the residual component, Represents the average trend of wind and solar power generation power signal;

[0026] Step 2.3, repeat Step 2.1 and Step 2.2 N times, and get The collection is:

[0027] ;

[0028] ;

[0029] S2.4, the corresponding Performing ensemble average operation, the final component of the ensemble empirical mode decomposition EEMD is obtained as:

[0030] ;

[0031] ;

[0032] In the above formula The first one represents the power load signal indivual Quantity; The first one represents the wind and solar power generation power signal indivual Quantity;

[0033] S2.5. Build an LSTM network to train each component signal;

[0034] S2.6. In LSTM networks, For the The component signal data input at the moment, is the cell state at the previous moment, where the initial state is set to 0. is the LSTM cell output at the previous moment. These three quantities are all passed into the cell. The specific calculation formula is as follows:

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] In the above formula, 、 、 Represent the forget gate, input gate and output gate respectively; and tanh function are activation functions; W and b are related coefficient matrices and biases;

[0043] S2.7. Introduce the GWO algorithm to optimize LSTM parameters. The GWO algorithm automatically optimizes network parameters by simulating the ecological phenomenon of wolf packs hunting. The wolf pack consists of four levels of wolves: Alpha, Beta, Delta, and Omega. Alpha is the only leader of the pack, and the pack's hunting process is divided into three stages:

[0044] S2.7.1. Encirclement: Under the leadership of the Alpha wolf, the wolf pack first encircles the prey. The mathematical model is as follows:

[0045] ;

[0046] ;

[0047] In the above formula, the following table i represents the number of iterations, ; D is the distance between the gray wolf and the prey; The number of iterations is i The prey position vector at time ; The number of iterations is i The gray wolf's position vector at time ; The number of iterations is i Gray wolf position vector at time +1; A andC is the coefficient vector, A and C The expression is:

[0048] ;

[0049] ;

[0050] In the above formula and is a random vector whose elements are in the range [0, 1]; a is a control parameter that decreases linearly from 2 to 0 as the number of iterations increases;

[0051] S2.7.2, Hunting; After the wolf pack surrounds the prey, the Alpha, Beta, and Delta wolves are closest to the prey. Led by the Alpha, Beta, and Delta wolves, the wolf pack will approach the prey. The position of the Omega wolf is updated as follows:

[0052] ;

[0053] ;

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] In the above formula 、 、 The distances between the current Omega wolf and the Alpha wolf, Beta wolf, and Delta wolf respectively; 、 、 are the position vectors of Alpha wolf, Beta wolf, and Delta wolf respectively;

[0060] S2.7.3, attack; the goal of this stage is to capture the prey, that is, to obtain the optimal solution for the LSTM network parameters; the GWO algorithm simulates the process of approaching the prey by reducing the value of a; when a gradually decreases, the element value of A is also in the range of [-a, a]. When , the wolves can attack the prey, that is, the LSTM network parameters find the optimal solution;

[0061] S2.8. Use the WGO algorithm to optimize the parameters of the LSTM network, including the number of hidden layer nodes, learning rate, and number of iterations.

[0062] The data prediction process based on ensemble empirical mode decomposition (EEMD) and the Gray Wolf-LSTM network includes adding white noise, performing empirical mode decomposition (EMD), ensemble averaging, building an LSTM network for training, and introducing the Gray Wolf Optimization (GWO) algorithm to optimize the LSTM parameters. This can more accurately capture the nonlinear characteristics of the data, improve prediction accuracy, automatically optimize the parameters of the LSTM network, and enhance the generalization ability of the model.

[0063] The real-time power load forecast value in S3 above Wind and photovoltaic power generation for:

[0064] ;

[0065] .

[0066] The power balance formula of the power grid in S4 above is:

[0067] + - =0

[0068] in, , is the total number of generator sets, To put the fixed power part of the generator set into operation, To put the frequency regulation power part of the generator set into operation, is the number of generating units put into operation, , is the tie line power, is the real-time power of electrochemical energy storage, When is the discharge state, It is in charging state.

[0069] A power grid power balance formula based on a generative adversarial network (GAN) is proposed. This formula combines the power load forecast value, wind and solar power generation forecast value, and the charge and discharge status of electrochemical energy storage to achieve power balance in the power grid. This power balance formula is used as a condition for subsequent adversarial network generation.

[0070] The single-cycle economic benefit objective function of electrochemical energy storage in S5 above is:

[0071] ;

[0072] Where T is the measurement period, , is the real-time power during discharge, When the electrochemical energy storage is not in the discharge state, the value is 0. The off-grid electricity price during discharge, calculated based on the coal-fired power price; , The peak electricity price is , The electricity price is flat. , The electricity price is in the off-peak period. 、 and The real-time charging power during peak, flat and valley periods respectively. When not in charging state during peak, flat and valley periods, the three values ​​are 0; is the capacity compensation income, For transmission and distribution costs, Other costs include equipment maintenance costs, operation and maintenance costs, and capacity electricity charges;

[0073] As the use cycle of electrochemical energy storage increases, the internal resistance of the battery pack increases, and the charging and discharging time is prolonged, which reduces the economic benefits of a single cycle and thus affects the investment returns of electrochemical energy storage.

[0074] From the perspective of maximizing the economic benefits of electrochemical investment, the goal is to maximize the profit per unit capacity in the long term. The long-term profit objective function is for:

[0075] ;

[0076] in For the benefits of electrochemical energy storage during its life cycle, is the number of metering cycles of the electrochemical energy storage life cycle, Q is the electrochemical energy storage capacity;

[0077] Long-term net profit objective function for:

[0078] ;

[0079] in is the investment cost, C is the investment cost of energy storage per unit capacity; is the electrochemical battery degradation loss, The coefficient of the loss of battery life-cycle benefits;

[0080] From the perspective of grid stability and long-term benefits of power plants, the single-cycle startup time of each generator set is The average generator unit rate V is a long-term guarantee for high power supply quality. The average generator unit rate V is:

[0081]

[0082] in, is the average number of generator sets started in a single cycle, The real-time number of boots;

[0083] Single cycle start time of each generator set and the average generator unit rate V and the long-term net profit objective function To ensure the mutual constraint relationship, weights are set for the two and the overall objective function R is obtained:

[0084] ;

[0085] in, and is the long-term net profit objective function and the weight of the average generator unit rate V. When the value of the objective function R is the set value, the long-term net profit objective function The relationship between the average number of generator sets V and the increase and decrease of the two is one of increase and decrease.

[0086] This paper describes in detail the construction process of the single-cycle economic benefit objective function, the long-cycle net income objective function, and the grid quality evaluation formula for electrochemical energy storage, and establishes an overall objective function model to balance economic benefits and grid quality. By constructing the economic benefit objective function and the grid quality evaluation formula, and establishing the overall objective function model, dual optimization of economic benefits and grid quality is achieved. The overall objective function model provides a clear optimization target for the operation and control of electrochemical energy storage power stations, helping to improve decision-making efficiency.

[0087] In the above S5, the generator is used to output the prediction scenario of the total objective function, generate scenarios and perform scenario reduction as follows:

[0088] The original data is used as training samples to train the generator and discriminator of the generative adversarial network. Historical data consistent with the current day's scenario is extracted, and the trained generator is used to output multiple prediction scenarios. The trained discriminator is then used to correct the prediction scenarios:

[0089] The generated prediction scenarios are divided into groups according to similarity, and one scenario in each group is selected as a typical scenario. The sum of the reduced scenario probabilities is 1. K scenarios are randomly selected from the generated scenarios as the initial cluster centers. According to the principle of the closest distance to the cluster center, the objects are assigned to each category. According to the minimum distance principle, a new cluster center is found to replace the original cluster center.

[0090] The constraints in S6 above include:

[0091] Average generator unit rate constraint: ;

[0092] and are the minimum and maximum values ​​for normal operation of the power plant, respectively;

[0093] Long-term net profit objective function constraints:

[0094] ;

[0095] and are the minimum long-period net benefit and the maximum long-period net benefit of the electrochemical energy storage power station respectively;

[0096] Capacity constraints of electrochemical energy storage stations:

[0097] ;

[0098] in, is the mean value of the power load, k is the ratio of the capacity of the electrochemical energy storage station to the mean value of the power load, and are the minimum and maximum values ​​of the ratio;

[0099] Charge and discharge approximately:

[0100] ;

[0101] in, is the ratio of discharge capacity to charge capacity in a single cycle, Single cycle The minimum value of .

[0102] The present invention provides an electrochemical energy storage power station operation control method suitable for spot trading. Based on historical power grid data, the LSMT and Gray Wolf algorithms are used to perform real-time predictions of wind and solar power output and power load during daytime trading. The objective function of the long-term profit of the electrochemical energy storage power station and the power grid stability index is established through a generative adversarial network. The optimal control method of the electrochemical energy storage power station is found through power balance control and power grid constraints. This method ensures improved charging and discharging stability of the electrochemical energy storage power station, ensures power supply quality of the power grid, optimizes the operating efficiency of the power plant power units, maximizes economic benefits, achieves real-time dynamic balance between the supply and demand of the electrochemical energy storage power station and the power grid, and improves the absorption rate of new energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] The present invention will be further described below with reference to the accompanying drawings and examples:

[0104] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0105] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0106] like Figure 1 As shown, the operation control method of the electrochemical energy storage power station applicable to spot trading includes the following steps:

[0107] S1. Collection and processing of raw data; raw data includes historical wind and solar power generation, historical power load, peak and valley metering periods, number and characteristic parameters of generator sets, and number and capacity of electrochemical energy storage groups;

[0108] S2, based on the historical power load data and historical wind and solar power generation in S1, uses ensemble empirical mode decomposition to decompose the data, and uses the Grey Wolf-LSMT network to train the decomposed signal;

[0109] S3: Use the model trained in S2 to predict and correct the current power load to obtain the power load prediction value of each component. Wind and photovoltaic power generation , the predicted values ​​of each component signal are superimposed to obtain the real-time power load prediction value Wind and photovoltaic power generation and make corrections;

[0110] S4. Based on the generative adversarial network, the power balance formula of the power grid is established by combining the power load forecast value in S3, the wind and photovoltaic power generation power forecast value in S4, and the charge and discharge status of the electrochemical energy storage;

[0111] S5. Based on the balance formula, the single-cycle economic benefit objective function formula of electrochemical energy storage and the grid quality evaluation formula are established to obtain the long-cycle net benefit objective function. , for the long-term net profit objective function The total objective function model is obtained by setting weights for the grid stability evaluation formula, and the prediction scenarios of the total objective function are output by the generator to generate scenarios and perform scenario reduction.

[0112] S6. Establish constraints for wind and solar power generation and electrochemical energy storage control, and output optimal electrochemical energy storage power station operation control decisions.

[0113] In the above S1, the original data is collected by collecting data at intervals, and the processing is carried out by eliminating obvious outliers and filling in missing values.

[0114] The specific steps in the above S2 are:

[0115] S2.1. Processed power load data Add a standard normal distribution of white noise , historical wind and solar power generation capacity Add a standard normal distribution of white noise , the specific formula is as follows:

[0116] ;

[0117] ;

[0118] In the above formula Indicates the The white noise sequence added is Indicates the The white noise sequence added is Indicates the The additional noise signal of the experiment, Indicates the Additional noise signal for the experiment;

[0119] S2.2. New signal sequences Perform empirical mode decomposition EMD and get modal components and 1 residual component:

[0120] ;

[0121] in, for The first transient components, is the residual component, Represents the average trend of the power load signal;

[0122] For new signal sequences Perform empirical mode decomposition EMD and get modal components and 1 residual component:

[0123] ;

[0124] in, for The first transient components, is the residual component, Represents the average trend of wind and solar power generation power signal;

[0125] Step 2.3, repeat Step 2.1 and Step 2.2 N times, and get The collection is:

[0126] ;

[0127] ;

[0128] S2.4, the corresponding Performing ensemble average operation, the final component of the ensemble empirical mode decomposition EEMD is obtained as:

[0129] ;

[0130] ;

[0131] In the above formula The first one represents the power load signal indivual Quantity; The first one represents the wind and solar power generation power signal indivual Quantity;

[0132] S2.5. Build an LSTM network to train each component signal;

[0133] S2.6. In LSTM networks, For the The component signal data input at the moment, is the cell state at the previous moment, where the initial state is set to 0. is the LSTM cell output at the previous moment. These three quantities are all passed into the cell. The specific calculation formula is as follows:

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] In the above formula, 、 、 Represent the forget gate, input gate and output gate respectively; and tanh function are activation functions; W and b are related coefficient matrices and biases;

[0142] S2.7. Introduce the GWO algorithm to optimize LSTM parameters. The GWO algorithm automatically optimizes network parameters by simulating the ecological phenomenon of wolf packs hunting. The wolf pack consists of four levels of wolves: Alpha, Beta, Delta, and Omega. Alpha is the only leader of the pack, and the pack's hunting process is divided into three stages:

[0143] S2.7.1. Encirclement: Under the leadership of the Alpha wolf, the wolf pack first encircles the prey. The mathematical model is as follows:

[0144] ;

[0145] ;

[0146] In the above formula, the following table i represents the number of iterations, ; D is the distance between the gray wolf and the prey; The number of iterations is i The prey position vector at time ; The number of iterations is i The gray wolf's position vector at time ; The number of iterations is i Gray wolf position vector at time +1; A and C is the coefficient vector, A and C The expression is:

[0147] ;

[0148] ;

[0149] In the above formula and is a random vector whose elements are in the range [0, 1]; a is a control parameter that decreases linearly from 2 to 0 as the number of iterations increases;

[0150] S2.7.2, Hunting; After the wolf pack surrounds the prey, the Alpha, Beta, and Delta wolves are closest to the prey. Led by the Alpha, Beta, and Delta wolves, the wolf pack will approach the prey. The position of the Omega wolf is updated as follows:

[0151] ;

[0152] ;

[0153] ;

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] In the above formula 、 、 The distances between the current Omega wolf and the Alpha wolf, Beta wolf, and Delta wolf respectively; 、 、 are the position vectors of Alpha wolf, Beta wolf, and Delta wolf respectively;

[0159] S2.7.3, attack; the goal of this stage is to capture the prey, that is, to obtain the optimal solution for the LSTM network parameters; the GWO algorithm simulates the process of approaching the prey by reducing the value of a; when a gradually decreases, the element value of A is also in the range of [-a, a]. When , the wolves can attack the prey, that is, the LSTM network parameters find the optimal solution;

[0160] S2.8. Use the WGO algorithm to optimize the parameters of the LSTM network, including the number of hidden layer nodes, learning rate, and number of iterations.

[0161] The real-time power load forecast value in S3 above Wind and photovoltaic power generation for:

[0162] ;

[0163] .

[0164] The power balance formula of the power grid in S4 above is:

[0165] + - =0

[0166] in, , is the total number of generator sets, To put the fixed power part of the generator set into operation, To put the frequency regulation power part of the generator set into operation, is the number of generating units put into operation, , is the tie line power, is the real-time power of electrochemical energy storage, When is the discharge state, It is in charging state.

[0167] The single-cycle economic benefit objective function of electrochemical energy storage in S5 above is:

[0168] ;

[0169] Where T is the measurement period, , is the real-time power during discharge, When the electrochemical energy storage is not in the discharge state, the value is 0. The off-grid electricity price during discharge, calculated based on the coal-fired power price; , The peak electricity price is , The electricity price is flat. , The electricity price is in the off-peak period. 、 and The real-time charging power during peak, flat and valley periods respectively. When not in charging state during peak, flat and valley periods, the three values ​​are 0; is the capacity compensation income, For transmission and distribution costs, Other costs include equipment maintenance costs, operation and maintenance costs, and capacity electricity charges;

[0170] As the use cycle of electrochemical energy storage increases, the internal resistance of the battery pack increases, and the charging and discharging time is prolonged, which reduces the economic benefits of a single cycle and thus affects the investment returns of electrochemical energy storage.

[0171] From the perspective of maximizing the economic benefits of electrochemical investment, the goal is to maximize the profit per unit capacity in the long term. The long-term profit objective function is for:

[0172] ;

[0173] in For the benefits of electrochemical energy storage during its life cycle, is the number of metering cycles of the electrochemical energy storage life cycle, Q is the electrochemical energy storage capacity;

[0174] Long-term net profit objective function for:

[0175] ;

[0176] in is the investment cost, C is the investment cost of energy storage per unit capacity; is the electrochemical battery degradation loss, The coefficient of the loss of battery life-cycle benefits;

[0177] From the perspective of grid stability and long-term benefits of power plants, the single-cycle startup time of each generator set is The average generator unit rate V is a long-term guarantee for high power supply quality. The average generator unit rate V is:

[0178]

[0179] in, is the average number of generator sets started in a single cycle, The real-time number of boots;

[0180] Single cycle start time of each generator set and the average generator unit rate V and the long-term net profit objective function To ensure the mutual constraint relationship, weights are set for the two and the overall objective function R is obtained:

[0181] ;

[0182] in, and is the long-term net profit objective function and the weight of the average generator unit rate V. When the value of the objective function R is the set value, the long-term net profit objective function The relationship between the average number of generator sets V and the increase and decrease of the two is one of increase and decrease.

[0183] In the above S5, the generator is used to output the prediction scenario of the total objective function, generate scenarios and perform scenario reduction as follows:

[0184] The original data is used as training samples to train the generator and discriminator of the generative adversarial network. Historical data consistent with the current day's scenario is extracted, and the trained generator is used to output multiple prediction scenarios. The trained discriminator is then used to correct the prediction scenarios:

[0185] The generated prediction scenarios are divided into groups according to similarity, and one scenario in each group is selected as a typical scenario. The sum of the reduced scenario probabilities is 1. K scenarios are randomly selected from the generated scenarios as the initial cluster centers. According to the principle of the closest distance to the cluster center, the objects are assigned to each category. According to the minimum distance principle, a new cluster center is found to replace the original cluster center.

[0186] The constraints in S6 above include:

[0187] Average generator unit rate constraint: ;

[0188] and are the minimum and maximum values ​​for normal operation of the power plant, respectively;

[0189] Long-term net profit objective function constraints:

[0190] ;

[0191] and The minimum and maximum long-term net benefits of an electrochemical energy storage power station are respectively. When the net benefit is lower than the minimum long-term net benefit, the electrochemical energy storage station cannot obtain sustainable investment. When the net benefit is higher than the maximum long-term net benefit, it will cause disorderly growth of the power station.

[0192] Capacity constraints of electrochemical energy storage stations:

[0193] ;

[0194] in, is the mean value of the power load, k is the ratio of the capacity of the electrochemical energy storage station to the mean value of the power load, and are the minimum and maximum values ​​of the ratio;

[0195] Charge and discharge approximately:

[0196] ;

[0197] in, is the ratio of discharge capacity to charge capacity in a single cycle, Single cycle The minimum value of .

Claims

1. An electrochemical energy storage power station operation control method suitable for spot trading, characterized in that: The following steps are involved: S1. Collection and processing of raw data; raw data includes historical wind and solar power generation, historical power load, peak and valley metering periods, number and characteristic parameters of generator sets, and number and capacity of electrochemical energy storage groups; S2, based on the historical power load data and historical wind and solar power generation in S1, uses ensemble empirical mode decomposition to decompose the data, and uses the Grey Wolf-LSMT network to train the decomposed signal; S3: Use the model trained in S2 to predict and correct the current power load to obtain the power load prediction value of each component. Wind and photovoltaic power generation , the predicted values ​​of each component signal are superimposed to obtain the real-time power load prediction value Wind and photovoltaic power generation and make corrections; S4. Based on the generative adversarial network, the power balance formula of the power grid is established by combining the power load forecast value in S3, the wind and photovoltaic power generation power forecast value in S4, and the charge and discharge status of the electrochemical energy storage; S5. Based on the balance formula, the single-cycle economic benefit objective function formula of electrochemical energy storage and the grid quality evaluation formula are established to obtain the long-cycle net benefit objective function. , for the long-term net profit objective function The total objective function model is obtained by setting weights for the grid stability evaluation formula, and the prediction scenario of the total objective function is output by the generator to generate scenarios and perform scenario reduction. The single-cycle economic benefit objective function of electrochemical energy storage is: ； Where T is the measurement period, , is the real-time power during discharge, When the electrochemical energy storage is not in the discharge state, the value is 0. The off-grid electricity price during discharge, calculated based on the coal-fired power price; , The peak electricity price is , The electricity price is flat. , The electricity price is in the off-peak period. 、 and The real-time charging power during peak, flat and valley periods respectively. When not in charging state during peak, flat and valley periods, the three values ​​are 0; is the capacity compensation income, For transmission and distribution costs, Other costs include equipment maintenance costs, operation and maintenance costs, and capacity electricity charges; As the use cycle of electrochemical energy storage increases, the internal resistance of the battery pack increases, and the charging and discharging time is prolonged, which reduces the economic benefits of a single cycle and thus affects the investment returns of electrochemical energy storage. From the perspective of maximizing the economic benefits of electrochemical investment, the goal is to maximize the profit per unit capacity in the long term. The long-term profit objective function is for: ； in For the benefits of electrochemical energy storage during its life cycle, is the number of metering cycles of the electrochemical energy storage life cycle, Q is the electrochemical energy storage capacity; Long-term net profit objective function for: ； in is the investment cost, C is the investment cost of energy storage per unit capacity; is the electrochemical battery degradation loss, The coefficient of the loss of battery life-cycle benefits; From the perspective of grid stability and long-term benefits of power plants, the single-cycle startup time of each generator set is The average generator unit rate V is a long-term guarantee for high power supply quality. The average generator unit rate V is: in, is the average number of generator sets started in a single cycle, The real-time number of boots; Single cycle start time of each generator set and the average generator unit rate V and the long-term net profit objective function To ensure the mutual constraint relationship, weights are set for the two and the overall objective function R is obtained: ； in, and is the long-term net profit objective function and the weight of the average generator unit rate V. When the value of the objective function R is the set value, the long-term net profit objective function The relationship between the average number of generator sets V and the increase and decrease is one of growth and loss; S6. Establish constraints for wind and solar power generation and electrochemical energy storage control, and output optimal electrochemical energy storage power station operation control decisions.

2. The electrochemical energy storage power station operation control method suitable for spot trading according to claim 1, characterized in that: In the above-mentioned S1, the original data is collected by collecting data at intervals, and the processing is performed by eliminating obvious outliers and filling in missing values.

3. The electrochemical energy storage power station operation control method suitable for spot trading according to claim 2, characterized in that: The specific steps in S2 are: S2.

1. Processed power load data Add a standard normally distributed white noise , historical wind and solar power generation capacity Add a standard normally distributed white noise , the specific formula is as follows: ； ； In the above formula Indicates the The white noise sequence added is Indicates the The white noise sequence added is Indicates the The additional noise signal of the experiment, Indicates the Additional noise signal for the experiment; S2.

2. New signal sequences Perform empirical mode decomposition EMD and get modal components and 1 residual component: ; in, for The first transient components, is the residual component, Represents the average trend of the power load signal; For new signal sequences Perform empirical mode decomposition EMD and get modal components and 1 residual component: ; in, for The first transient components, is the residual component, Represents the average trend of wind and solar power generation power signal; S2.3, repeat Step 2.1 and Step 2.2 N times, and get The collection is: ； ； S2.4, the corresponding Performing ensemble average operation, the final component of the ensemble empirical mode decomposition EEMD is obtained as: ； ； In the above formula The first one represents the power load signal indivual Quantity; The first one represents the wind and solar power generation power signal indivual Quantity; S2.

5. Build an LSTM network to train each component signal; S2.

6. In LSTM networks, For the The component signal data input at the moment, is the cell state at the previous moment, where the initial state is set to 0. is the LSTM cell output at the previous moment. These three quantities are all passed into the cell. The specific calculation formula is as follows: ； ； ； ； ； ； ； In the above formula, 、 、 Represent the forget gate, input gate and output gate respectively; and tanh function are activation functions; W and b are related coefficient matrices and biases; S2.

7. Introduce the GWO algorithm to optimize LSTM parameters. The GWO algorithm automatically optimizes network parameters by simulating the ecological phenomenon of wolf pack hunting. The wolf pack consists of four levels of wolves: Alpha, Beta, Delta, and Omega. Alpha is the sole leader of the pack, and the pack's hunting process is divided into three stages: S2.7.

1. Encirclement: Under the leadership of the Alpha wolf, the wolf pack first encircles the prey. The mathematical model is as follows: ； ； In the above formula, the following table i represents the number of iterations, ; D is the distance between the gray wolf and the prey; The number of iterations is i The prey position vector at time ; The number of iterations is i The gray wolf's position vector at time ; The number of iterations is i Gray wolf position vector at time +1; A and C is the coefficient vector, A and C The expression is: ； ； In the above formula and is a random vector whose elements are in the range [0, 1]; a is a control parameter that decreases linearly from 2 to 0 as the number of iterations increases; S2.7.2, Hunting; After the wolf pack surrounds the prey, the Alpha, Beta, and Delta wolves are closest to the prey. Led by the Alpha, Beta, and Delta wolves, the wolf pack will approach the prey. The position of the Omega wolf is updated as follows: ； ； ； ； ； ； ； In the above formula 、 、 The distances between the current Omega wolf and the Alpha wolf, Beta wolf, and Delta wolf respectively; 、 、 are the position vectors of Alpha wolf, Beta wolf, and Delta wolf respectively; S2.7.3, attack; the goal of this stage is to capture the prey, that is, to obtain the optimal solution for the LSTM network parameters; the GWO algorithm simulates the process of approaching the prey by reducing the value of a; when a gradually decreases, the element value of A is also in the range of [-a, a]. When , the wolves can attack the prey, that is, the LSTM network parameters find the optimal solution; S2.

8. Use the WGO algorithm to optimize the parameters of the LSTM network, including the number of hidden layer nodes, learning rate, and number of iterations.

4. The electrochemical energy storage power station operation control method suitable for spot trading according to claim 3 is characterized in that: The real-time power load forecast value in S3 Wind and photovoltaic power generation for: ； 。 5. The electrochemical energy storage power station operation control method suitable for spot trading according to claim 4 is characterized in that: The power balance formula of the power grid in S4 is: + - =0 in, , is the total number of generator sets, To put the fixed power part of the generator set into operation, To put the frequency regulation power part of the generator set into operation, is the number of generating units put into operation, , is the tie line power, is the real-time power of electrochemical energy storage, When is the discharge state, It is in charging state.

6. The electrochemical energy storage power station operation control method suitable for spot trading according to claim 5, characterized in that: In the above S5, the generator is used to output the prediction scenario of the total objective function, generate the scenario and perform scenario reduction as follows: The original data is used as training samples to train the generator and discriminator of the generative adversarial network. Historical data consistent with the current day's scenario is extracted, and the trained generator is used to output multiple prediction scenarios. The trained discriminator is then used to correct the prediction scenarios: The generated prediction scenarios are divided into groups according to similarity, and one scenario in each group is selected as a typical scenario. The sum of the reduced scenario probabilities is 1. K scenarios are randomly selected from the generated scenarios as the initial cluster centers. According to the principle of the closest distance to the cluster center, the objects are assigned to each category. According to the minimum distance principle, a new cluster center is found to replace the original cluster center.

7. The electrochemical energy storage power station operation control method suitable for spot trading according to claim 6, characterized in that: The constraints in S6 include: Average generator unit rate constraint: ; and are the minimum and maximum values ​​for normal operation of the power plant, respectively; Long-term net profit objective function constraints: ； and are the minimum long-period net benefit and the maximum long-period net benefit of the electrochemical energy storage power station respectively; Capacity constraints of electrochemical energy storage stations: ； in, is the mean value of the power load, k is the ratio of the capacity of the electrochemical energy storage station to the mean value of the power load, and are the minimum and maximum values ​​of the ratio; Charge and discharge approximately: ； in, is the ratio of discharge capacity to charge capacity in a single cycle, Single cycle The minimum value of .

Images