Frequency modulation income optimization method for independent energy storage system

Abstract

The invention discloses a frequency modulation income optimization method for an independent energy storage system. The method comprises the following steps: obtaining a power grid operation data set and a market transaction data set and carrying out format standardization; constructing a time sequence data set based on the standardized power grid operation data set and the market transaction data set; constructing and training a one-dimensional convolution gating circulation network model by using the time sequence data set, and generating a prediction result of a future frequency modulation load and a market electricity price; according to a prediction result, constructing a frequency modulation optimization problem taking frequency modulation economic benefit maximization as a target, and generating a frequency modulation optimization strategy space containing a plurality of independent energy storage frequency modulation strategies; performing iterative search on the frequency modulation optimization strategy space to obtain an optimal independent energy storage frequency modulation strategy; and controlling the energy storage equipment to perform frequency modulation operation according to the optimal independent energy storage frequency modulation strategy. According to the scheme, high prediction stability and response capability can be kept under the market conditions of high volatility and frequent change of electricity price.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management of energy storage systems, and in particular to a method for optimizing frequency modulation revenue for an independent energy storage system. Background Art

[0002] With the continuous deepening of the electricity market mechanism and the continuous development of renewable energy, the role of independent energy storage systems in power frequency modulation has become increasingly prominent. As an important tool for regulating the power grid frequency, alleviating supply-demand fluctuations, and enhancing the consumption capacity of new energy, independent energy storage systems are widely used in the electricity trading and ancillary service markets.

[0003] At present, the frequency modulation strategies of independent energy storage systems mostly rely on rule-driven or simple linear optimization models. These methods have obvious limitations when facing a power market environment with multi-factor coupling, rapid state changes, and high nonlinearity. On the one hand, traditional models are difficult to dynamically perceive multi-source information such as market prices, grid frequencies, and load changes, and the frequency modulation strategies lack real-time and adaptive capabilities, resulting in sub-optimal revenue. On the other hand, existing methods mostly adopt static parameter settings or fixed scheduling logics, and are unable to extract hidden time-series features and revenue-driven laws based on historical operation data, and the strategies often have lag and local optimum problems.

[0004] Although some studies have introduced machine learning and intelligent optimization algorithms in recent years to try to improve the prediction accuracy and flexibility of scheduling strategies, most methods have problems such as single training objectives, prediction structures that do not have the ability to model long-term dependencies, and optimization algorithms falling into local optima, and are difficult to effectively support the goal of maximizing revenue in the electricity trading environment. At the same time, existing models generally lack operation feedback and scheduling closed-loop mechanisms, and are unable to achieve real-time monitoring and dynamic adjustment of the execution effects of strategies, resulting in poor adaptability of the scheduling model to actual power system fluctuations.

[0005] In summary, there is an urgent need for a new method that combines prediction modeling and intelligent optimization, and takes into account both profitability and operation safety to break through the current technical bottlenecks. Summary of the Invention

[0006] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further described in detail in the Detailed Implementation section. The Summary of the Invention section of the present invention does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the protection scope of the claimed technical solution.

[0007] To at least partially solve the above problems, the present invention provides a method for optimizing frequency modulation revenue for an independent energy storage system, including:

[0008] S1. Obtain the power grid operation dataset and the market transaction dataset and standardize their formats;

[0009] S2. Based on the standardized power grid operation dataset and market transaction dataset, construct a time series dataset;

[0010] S3. Use the time series dataset to construct and train a one-dimensional convolutional gated recurrent network model to generate prediction results for future frequency regulation load and market electricity price;

[0011] S4. According to the prediction results, construct a frequency regulation optimization problem aiming at maximizing the frequency regulation economic benefit, including the operation range of the energy storage system, the upper limit of energy storage capacity, system stability constraints and market transaction rule constraints, and generate a frequency regulation optimization strategy space containing multiple independent energy storage frequency regulation strategies;

[0012] S5. Through iterative search of the frequency regulation optimization strategy space, obtain the optimal independent energy storage frequency regulation strategy;

[0013] S6. Control the energy storage device to perform frequency regulation operations according to the optimal independent energy storage frequency regulation strategy.

[0014] Further, the power grid operation data in step S1 includes power grid frequency, power load, market electricity price and energy storage system status data, and the market transaction data includes the transaction volume of power transactions.

[0015] Further, step S2 includes:

[0016] S21. Align the standardized power grid operation dataset and market transaction dataset according to a unified time step, and sequentially construct a multi-dimensional time series feature vector sequence sorted by time, where the multi-dimensional time series feature vector sequence includes power grid operation data and market transaction data at each time step;

[0017] S22. In the multi-dimensional time series feature vector sequence, set a sliding time window with a fixed length, and divide the time series data into multiple consecutive sliding window subsequences in a sliding window manner. Each sliding window subsequence contains data of a fixed number of consecutive time steps, which is used to establish a model of the change process of the power system state in the time dimension, where the data dimension of each sliding window subsequence is equal to the product of the window length and the feature dimension;

[0018] S23. Divide each sliding window subsequence into an input part and a target prediction part. The input part consists of the feature vectors of several time steps before this sliding window subsequence and is used as the input for training the one-dimensional convolutional gated recurrent network model. The target prediction part is the feature vector of the last time step of this sliding window subsequence and is used as the output for training the one-dimensional convolutional gated recurrent network model, representing the future power grid operation state and market price behavior that the one-dimensional convolutional gated recurrent network model needs to predict. Then, construct a training sample set according to the division rule for all sliding window subsequences.

[0019] Further, the training sample set consists of sample pairs composed of the input part and the corresponding target prediction part:

[0020]

[0021] Among them, represents the input part in the k-th sliding window subsequence, which consists of multi-dimensional time series feature vectors of the first L w -1 time steps. The dimension of the multi-dimensional time series feature vector is 5, corresponding to the power grid frequency, power load, market electricity price, state of charge of the energy storage system, and electricity trading volume. Y k represents the target prediction part in the k-th sliding window subsequence, and its dimension is the same as that of the feature vector of each time step in the input part of the sliding window subsequence . T train represents the training sample set. k ∈ [1, T - L w +1] represents the training sample index range. T is the total number of time steps, and L w is the sliding window length, that is, the number of time steps for prediction within the sliding window.

[0022] Further, the step S4 includes:

[0023] S41. Based on the trained one-dimensional convolutional gated recurrent network model, predict the power grid operation state and market price behavior of multiple future time steps, and construct a prediction input set. The prediction input set includes the predicted values of the power grid frequency, power load, market electricity price, state of charge of the energy storage system, and electricity trading volume for each future time step;

[0024] S42. Based on the predicted value of the market electricity price and the scheduled discharge power of the energy storage system in the prediction input set, combine the capacity decay cost caused by the use of the energy storage battery to construct the frequency modulation economic benefit function R profit :

[0025]

[0026] Among them, is the predicted value of the market electricity price, represents the dispatching discharge power of the energy storage system, is the capacity attenuation cost caused by the use of the energy storage battery;

[0027] S43. Based on the predicted values of the state of charge of the energy storage system, the predicted value of the grid frequency, the predicted value of the electricity trading volume, and the dispatching discharge power of the energy storage system in the predicted input set, establish a frequency modulation optimization constraint set;

[0028] S44. With the frequency modulation economic benefit as the optimization objective and the frequency modulation optimization constraint set as the feasible solution space, construct a frequency modulation optimization strategy space function S opt :

[0029]

[0030] wherein, is the frequency modulation optimization constraint set, represents the predicted input set.

[0031] Furthermore, the frequency modulation optimization constraint set includes:

[0032] The state of charge boundary constraint of the energy storage system requires that the state of charge of the energy storage system at each future time step is always between the preset minimum state of charge and the maximum state of charge;

[0033] The power limit constraint requires that the dispatching value of the energy storage power at each time step does not exceed its maximum discharge power and maximum charge power;

[0034] The energy balance constraint requires that the state of charge at each time step is updated according to the state of charge of the previous step, the battery efficiency, the power value, and the energy storage capacity;

[0035] The frequency and market compliance constraint requires that the deviation of the predicted value of the grid frequency does not exceed the specified frequency tolerance range, and the product of the dispatching power of the energy storage system and the market trading volume at any time step does not exceed the maximum trading capacity limit allowed by the market.

[0036] Furthermore, the step S5 includes:

[0037] S51. Initialize the firefly optimization algorithm population based on the frequency modulation optimization strategy space function, and each firefly individual corresponds to an independent energy storage frequency modulation strategy P in the frequency modulation optimization strategy space k ;

[0038] S52. For each independent energy storage frequency modulation strategy P k , calculate its frequency modulation economic benefit as the fitness function value according to the predicted input set :

[0039]

[0040] Among them, represents the discharge power part corresponding to the k-th independent energy storage frequency modulation strategy at the t-th time step, represents the capacity attenuation cost caused by the energy storage discharge behavior of the k-th independent energy storage frequency modulation strategy at the t-th time step;

[0041] S53. Based on the fitness function value F k , according to the luminous intensity principle of the firefly optimization algorithm, define the luminous intensity I K of the k-th firefly individual as its frequency modulation economic benefit R profit (P k ), and calculate the mutual attraction β kj between firefly individuals according to the attraction update rule;

[0042] S54. Perform a position update operation on all independent energy storage frequency modulation strategies in each round of iteration, and update the frequency modulation strategy by combining the attraction mechanism and the local perturbation mechanism:

[0043]

[0044] Among them, P j (t) is the independent energy storage frequency modulation strategy with higher fitness relative to the firefly individual k in the t-th round, ∈ is the local random perturbation factor, and rand() represents a random variable obeying the uniform distribution on [0, 1];

[0045] S55. After each position update, perform a feasibility check on the updated independent energy storage frequency modulation strategy under the described frequency modulation optimization constraint set , including the state of charge boundary, power boundary, energy balance, frequency deviation, and market trading capacity upper limit constraints of the energy storage system. If any of the constraints is violated, roll back and correct the illegal time step or randomly resample, so that each strategy firefly individual is legal in the frequency modulation optimization strategy space;

[0046] S56. Repeat steps S52 - S55 until the maximum number of iterations T max or the global fitness value of the population meets the convergence criterion, and output the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy with the maximum frequency modulation benefit in the current population:

[0047]

[0048] Furthermore, the step S6 includes:

[0049] S61. The energy storage power scheduling value P corresponding to the optimal independent energy storage frequency modulation strategy* Applied to the power trading system to represent the energy storage power scheduling value of the optimal independent energy storage frequency modulation strategy at the t-th time step. A positive value indicates discharging, and a negative value indicates charging. The system sends the energy storage power scheduling value of each time step to the energy storage control module as a control instruction;

[0050] S62. The energy storage system control unit receives and parses the control instruction of each time step, and drives the energy conversion module and the battery pack in the energy storage device to perform corresponding charge and discharge operations, realizing hourly frequency modulation of the independent energy storage device. At the same time, the control instruction is connected to the power market management system to realize coordinated response with the frequency modulation market instruction;

[0051] S63. During the process of the energy storage device performing frequency modulation operation, multiple system operation state parameters are collected in real time to form an operation monitoring index set, including the actual response power, that is, the real output power of the energy storage system at the t-th time step, the state of charge of the energy storage system, that is, the SOC value at the t-th time step, the actual frequency modulation revenue obtained, that is, the real-time revenue calculated according to the market electricity price and the energy storage scheduling behavior, the grid frequency response value, that is, the grid frequency measured at the t-th time step, and the market transaction electricity price, that is, the electricity price at which the energy storage system participates in frequency modulation trading at the t-th time step;

[0052] S64. Conduct deviation analysis on the operation monitoring index set collected in real time, compare the execution effect of the optimal independent energy storage frequency modulation strategy with the actual operation state step by step in time, and calculate the deviation index;

[0053] S65. If the cumulative result of the deviation index exceeds the set threshold, the system will judge that there is a performance decline or strategy deviation in the execution of energy storage scheduling, and then trigger an adaptive feedback mechanism, and send the execution deviation information including power deviation, state of charge deviation and revenue deviation back to the one-dimensional convolutional gated recurrent network model and the firefly optimization algorithm module;

[0054] S66. After the execution of the frequency modulation strategy is completed, the frequency modulation execution data of the whole cycle is structurally integrated to generate a frequency modulation execution report.

[0055] Furthermore, the deviation index in step S64 includes:

[0056] Power deviation, that is, the absolute value of the difference between the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy and the actual corresponding power;

[0057] State of charge deviation, that is, the difference between the actual state of charge of the energy storage system and the predicted state of charge;

[0058] The revenue deviation is the difference between the theoretical revenue calculated based on the actual market electricity price and the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy and the actual frequency modulation revenue.

[0059] Further, the frequency modulation execution report in step S66 includes the energy storage power scheduling value, the actual power response value, the state of charge change trajectory, the grid frequency response, the electricity market price, and the final frequency modulation revenue data corresponding to the optimal independent energy storage frequency modulation strategy at each time step.

[0060] The beneficial effects of the present invention are as follows:

[0061] (1) In the time series prediction model of the present invention, a multi-scale convolution structure is introduced to extract the power system change characteristics at different time scales, the gated recurrent unit is combined to model the long-term dependence relationship of electricity price, load and state of charge, and at the same time, a revenue weight vector is introduced to dynamically weight the importance of each feature in the frequency modulation revenue, guiding the network to focus on the key variables affecting economic benefits, realizing the deep coupling of the prediction output and the subsequent scheduling optimization goal, the structure is more target-oriented, and it can maintain high prediction stability and response ability under market conditions with strong volatility and frequent electricity price changes.

[0062] (2) In the optimization strategy modeling of the present invention, multiple actual operation constraints such as the state of charge boundary of the energy storage, the charge and discharge power limit, the energy balance equation, the grid frequency offset tolerance, and the market trading capacity constraint are systematically integrated. The defined frequency modulation optimization strategy space not only maintains the global feasibility of the solution space, but also can fully express the coupling relationship between market dynamics and equipment operation, significantly improving the consistency between the optimization goal and engineering practice, and avoiding the model-system disconnection problem existing in the existing methods.

[0063] (3) The present invention combines the light intensity mechanism in the traditional firefly algorithm with the energy storage frequency modulation economic revenue function, encodes the predicted market electricity price and the discharge strategy together as the fitness value of the objective function, and introduces a constraint verification and local perturbation mechanism in the individual update process, so that the algorithm is always controlled by the boundary constraints of the frequency modulation optimization strategy space during the search process, ensuring the legality of the solution and the executability of the system operation. The firefly algorithm of the present invention has stronger dynamic adaptability and search accuracy, and realizes the global optimum and stable convergence of the scheduling strategy in the frequency modulation scenario with multiple factors intertwined. Description of the Drawings

[0064] The following drawings of the present invention are used as a part of the present invention to understand the present invention. The embodiments and descriptions of the present invention are shown in the drawings to explain the device and principle of the present invention. In the drawings,

[0065] Figure 1 is a flowchart of the frequency modulation revenue optimization method for an independent energy storage system according to an embodiment of the present invention. Detailed implementation manners

[0066] In the following description, numerous specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other instances, in order to avoid confusion with the present invention, some well-known technical features are not described.

[0067] To thoroughly understand the present invention, detailed structures will be presented in the following description to illustrate the present invention. Obviously, the implementation of the present invention is not limited to the specific details familiar to those skilled in the art of this technology field. The preferred implementation manners of the present invention are described in detail below. However, in addition to these detailed descriptions, the present invention can also have other implementation manners and should not be construed as being limited to the implementation manners presented here.

[0068] It should be understood that the purpose of the terms used herein is only to describe specific implementation manners and is not a limitation of the present invention. The singular forms of "a", "an" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. When the terms "comprise" and / or "include" are used in this specification, they specify the presence of the described features, wholes, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations. The terms "upper", "lower", "front", "rear", "left", "right" and similar expressions used in the present invention are only for the purpose of illustration and are not limitations.

[0069] The ordinal numbers such as "first" and "second" cited in the present invention are only identifiers and do not have any other meanings, such as a specific order, etc. Moreover, for example, the term "first component" does not imply the existence of a "second component" by itself, and the term "second component" does not imply the existence of a "first component" by itself.

[0070] Hereinafter, the specific implementation manners of the present invention will be described in more detail with reference to the accompanying drawings, which show representative implementation manners of the present invention and do not limit the present invention.

[0071] The present invention provides a method for optimizing the frequency modulation revenue of an independent energy storage system, as Figure 1 shown, the method includes:

[0072] Step S1, obtaining a power grid operation data set and a market transaction data set and performing format standardization.

[0073] The above-mentioned power grid operation data may specifically include power grid frequency, power load, market electricity price and energy storage system status data, and the market transaction data includes the transaction volume of power trading.

[0074] Step S2: Based on the standardized power grid operation dataset and market transaction dataset, construct a time series dataset.

[0075] This step can specifically include:

[0076] S21: Align the standardized power grid operation dataset and market transaction dataset according to a unified time step, and sequentially construct a multi-dimensional time series feature vector sequence sorted by time. The multi-dimensional time series feature vector sequence includes power grid operation data and market transaction data at each time step;

[0077] S22: In the multi-dimensional time series feature vector sequence, set a sliding time window with a fixed length, and divide the time series data into multiple consecutive sliding window subsequences in a sliding window manner. Each sliding window subsequence contains data of a fixed number of consecutive time steps, which is used to establish a model for the change process of the power system state in the time dimension. Among them, the data dimension of each sliding window subsequence is equal to the product of the window length and the feature dimension;

[0078] S23: Divide each sliding window subsequence into an input part and a target prediction part. The input part is composed of the feature vectors of the first several time steps of the sliding window subsequence and is used as the training input of the one-dimensional convolutional gated recurrent network model. The target prediction part is the feature vector of the last time step of the sliding window subsequence and is used as the training output of the one-dimensional convolutional gated recurrent network model, indicating the future power grid operation state and market price behavior that the one-dimensional convolutional gated recurrent network model needs to predict. And all the sliding window subsequences are formed into a training sample set according to the division rules.

[0079] Step S3: Use the time series dataset to construct and train a one-dimensional convolutional gated recurrent network model to generate prediction results of future frequency modulation load and market electricity price.

[0080] The training sample set of the one-dimensional convolutional gated recurrent network model is all the sliding window subsequences described in step S2 above, where the training input is the input part of each sliding window subsequence, and the training output is the target prediction part of each sliding window subsequence. Specifically, the training sample set is composed of sample pairs consisting of the input part and the corresponding target prediction part:

[0081]

[0082] Among them, represents the input part in the k-th sliding window subsequence, which is composed of the first L in the sliding window subsequence w- It consists of multi-dimensional time series feature vectors for 1 time step. The dimension of the multi-dimensional time series feature vector is 5, corresponding to grid frequency, power load, market electricity price, state of charge of the energy storage system, and electricity trading volume, Y k represents the target prediction part in the k-th sliding window subsequence, and the dimension is the same as the input part in the sliding window subsequence for each time step in, T train represents the training sample set, k ∈ [1, T - L w +1] represents the training sample index range, T is the total number of time steps, L w is the sliding window length, that is, the number of time steps for prediction within the sliding window; Lw-1 above is a moment within the sliding window, representing the time step one step away from the last moment within the current time window, used to calibrate the input features of the previous time step, that is, at the previous moment of the window length Lw, as the data item input to the model.

[0083] Step S4. According to the prediction results, construct a frequency regulation optimization problem with the goal of maximizing the frequency regulation economic benefit, including the operating range of the energy storage system, the upper limit of energy storage capacity, system stability constraints, and market trading rule constraints, and generate a frequency regulation optimization strategy space containing multiple independent energy storage frequency regulation strategies.

[0084] This step can specifically include:

[0085] S41. Based on the trained one-dimensional convolutional gated recurrent network model, predict the grid operating state and market price behavior for multiple future time steps, and construct a prediction input set. The prediction input set includes the predicted values of grid frequency, power load, market electricity price, state of charge of the energy storage system, and electricity trading volume for each future time step (where the predicted values of grid frequency and market electricity price need to be based on the predicted value of power load);

[0086] S42. Based on the predicted value of the market electricity price in the prediction input set and the scheduling discharge power of the energy storage system, combine the capacity attenuation cost caused by the use of the energy storage battery to construct a frequency regulation economic benefit function R for the future time range profit :[[]]END]]

[0087]

[0088] Among them, is the predicted value of the market electricity price, represents the scheduling discharge power of the energy storage system, is the capacity attenuation cost caused by the use of the energy storage battery;

[0089] S43. Based on the predicted state of charge value of the energy storage system, the predicted power grid frequency value, the predicted electricity trading volume value, and the scheduled discharge power of the energy storage system in the predicted input set, establish a frequency regulation optimization constraint set;

[0090] S44. Taking the frequency regulation economic benefit as the optimization goal and the frequency regulation optimization constraint set as the feasible solution space, construct a frequency regulation optimization strategy space function S opt :

[0091]

[0092] wherein, is the frequency regulation optimization constraint set, represents the predicted input set.

[0093] The frequency regulation optimization constraint set in step S43 above includes:

[0094] The state of charge boundary constraint of the energy storage system requires that the state of charge of the energy storage system at each future time step is always between the preset minimum state of charge and the maximum state of charge;

[0095] The power limit constraint requires that the scheduled value of the energy storage power at each time step does not exceed its maximum discharge power and maximum charge power;

[0096] The energy balance constraint requires that the state of charge at each time step is updated according to the state of charge of the previous step, the battery efficiency, the power value, and the energy storage capacity;

[0097] The frequency and market compliance constraint requires that the deviation of the predicted power grid frequency value does not exceed the specified frequency tolerance range, and the product of the scheduled power of the energy storage system and the market trading volume at any time step does not exceed the maximum trading capacity limit allowed by the market.

[0098] Step S5. By performing iterative search on the frequency regulation optimization strategy space, obtain the optimal independent energy storage frequency regulation strategy.

[0099] This step may specifically include:

[0100] S51. Initialize the population of the firefly optimization algorithm based on the frequency regulation optimization strategy space function, and each firefly individual corresponds to an independent energy storage frequency regulation strategy P in the frequency regulation optimization strategy space k ;

[0101] S52. For each independent energy storage frequency regulation strategy P k , calculate its frequency regulation economic benefit as the fitness function value according to the predicted input set :

[0102]

[0103] Among them, represents the discharge power part corresponding to the k-th independent energy storage frequency modulation strategy at the t-th time step, represents the capacity attenuation cost caused by the energy storage discharge behavior of the k-th independent energy storage frequency modulation strategy at the t-th time step;

[0104] S53. Based on the fitness function value F k , according to the luminous intensity principle of the firefly optimization algorithm, define the luminous intensity I of the k-th firefly individual K as its frequency modulation economic benefit R profit (P k ), and calculate the mutual attraction β between firefly individuals according to the attraction update rule kj ;

[0105] S54. Perform position update operations on all independent energy storage frequency modulation strategies in each iteration, and update the frequency modulation strategy by combining the attraction mechanism and the local perturbation mechanism:

[0106]

[0107] Among them, P j (t) is the independent energy storage frequency modulation strategy with higher fitness relative to the firefly individual k in the t-th round, ∈ is the local random perturbation factor, and rand() represents a random variable subject to a uniform distribution on [0, 1];

[0108] S55. After each position update, perform feasibility verification on the updated independent energy storage frequency modulation strategy under the above-mentioned frequency modulation optimization constraint set , including the state of charge boundary, power boundary, energy balance, frequency offset, and market trading capacity upper limit constraints of the energy storage system. If any of the constraints is violated, roll back and correct the violation time step or randomly resample, so that each strategy firefly individual is legal within the frequency modulation optimization strategy space;

[0109] S56. Repeat steps S52 - S55 until the maximum number of iterations T max or the global fitness value of the population satisfies the convergence criterion, and output the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy with the maximum frequency modulation benefit in the current population:

[0110]

[0111] Step S6. Control the energy storage device to perform frequency modulation operations according to the optimal independent energy storage frequency modulation strategy.

[0112] This step specifically includes:

[0113] S61. Apply the energy storage power scheduling value \(P\) corresponding to the optimal independent energy storage frequency modulation strategy * to the power trading system to represent the energy storage power scheduling value of the optimal independent energy storage frequency modulation strategy at the \(t\)-th time step. A positive value indicates discharging, and a negative value indicates charging. The system will send the energy storage power scheduling value of each time step to the energy storage control module as a control instruction;

[0114] S62. The energy storage system control unit receives and parses the control instruction of each time step, and drives the energy conversion module and the battery pack in the energy storage device to perform corresponding charge and discharge operations to achieve hourly frequency modulation of the independent energy storage device. At the same time, the control instruction is connected to the power market management system to achieve coordinated response with the frequency modulation market instruction;

[0115] S63. During the process of the energy storage device performing frequency modulation operations, multiple system operation state parameters are collected in real time to form an operation monitoring index set, including the actual response power, that is, the true output power of the energy storage system at the \(t\)-th time step, the state of charge of the energy storage system, that is, the SOC value at the \(t\)-th time step, the actual frequency modulation revenue obtained, that is, the real-time revenue calculated according to the market electricity price and the energy storage scheduling behavior, the power grid frequency response value, that is, the power grid frequency measured at the \(t\)-th time step, and the market transaction electricity price, that is, the electricity price at which the energy storage system participates in frequency modulation transactions at the \(t\)-th time step;

[0116] S64. Conduct deviation analysis on the operation monitoring index set collected in real time, compare the execution effect of the optimal independent energy storage frequency modulation strategy with the actual operation state step by step in time, and calculate deviation indexes. The deviation indexes specifically include:

[0117] Power deviation, that is, the absolute value of the difference between the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy and the actual corresponding power;

[0118] State of charge deviation, that is, the difference between the actual state of charge of the energy storage system and the predicted state of charge;

[0119] Revenue deviation, that is, the difference between the theoretical revenue calculated according to the actual market electricity price and the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy and the actual frequency modulation revenue;

[0120] S65. If the cumulative result of the deviation indexes exceeds the set threshold, the system will judge that there is a performance decline or strategy deviation in the execution of energy storage scheduling, and then trigger an adaptive feedback mechanism to send the execution deviation information including power deviation, state of charge deviation and revenue deviation back to the one-dimensional convolutional gated recurrent network model and the firefly optimization algorithm module;

[0121] S66. After the frequency regulation strategy is executed, the frequency regulation execution data of the entire cycle is structured and integrated to generate a frequency regulation execution report. The frequency regulation execution report specifically includes the energy storage power dispatch value, actual power response value, charge state change trajectory, grid frequency response, electricity market price and final frequency regulation revenue data corresponding to the optimal independent energy storage frequency regulation strategy at each time step.

[0122] The following provides specific embodiments to illustrate the technical solution of the present invention:

[0123] In May 2024, a provincial power grid control center conducted a special test on the frequency control response mechanism under the background of a high proportion of renewable energy grid connection. The test object was Binzhou Independent Energy Storage Power Station A, which was equipped with a 100MW / 200MWh lithium battery system. The system has been connected to the electricity spot market and has the qualifications to participate in power auxiliary services. With the increase in summer load, regional wind power and photovoltaic output fluctuations have intensified, which has brought great pressure to frequency control. The price in the electricity trading market fluctuated violently, with the intraday high and low price difference reaching 0.52 yuan / kWh and the maximum frequency deviation reaching ±0.36Hz. There is an urgent need for a frequency control method that takes into account both the timeliness of frequency response and the optimization of economic benefits.

[0124] In order to verify the feasibility of the present invention, the grid side decided to compare and evaluate the control effects of the present invention and the traditional static frequency regulation strategy method. The test period was set to 7 days from May 10, 2024 to May 16, 2024, and the control effect was monitored from 9:00 am to 9:00 pm every day to meet the evaluation requirements of the peak period of the main load of frequency regulation.

[0125] First, a sample set is constructed based on the historical data of the power grid from June 2023 to April 2024, including the power grid frequency, regional load, spot market electricity price, energy storage charge state and actual transaction volume sampled every 5 minutes, totaling about 105,120 time series data samples. The one-dimensional convolutional gated recurrent network structure proposed in this invention is adopted, and the input sliding window length is set to 24 steps, corresponding to the system operation status in the past 2 hours. Each window sample contains five types of features: frequency, load, electricity price, charge state, and transaction volume, and the dimension of each type of feature is unified to 1.

[0126] In the network structure design, three groups of convolution kernels of different scales are used to extract local features of 10 minutes, 30 minutes and 60 minutes granularity respectively, and then a two-layer GRU structure is connected to extract deep temporal dependencies. Finally, the output is fused through the revenue-aware weight vector to output the grid frequency and electricity price forecast values for the next 6 time steps (i.e. 30 minutes). The number of training rounds is set to 100 rounds, the batch size is 64, the Adam optimizer is used, and the initial learning rate is 0.001. The model finally achieved a mean square error of 0.015 (yuan) in electricity price prediction on the validation set. 2 / kWh2 ) The frequency prediction error is 0.024 Hz, which is better than 0.033 and 0.038 of the traditional LSTM model.

[0127] After the prediction module is completed, based on the future electricity price and load prediction results, a frequency regulation optimization problem is constructed and the firefly optimization algorithm is called to search for strategies. The optimization goal is set to maximize the cumulative frequency regulation revenue in the next 30 minutes. The constraints include that the upper and lower limits of the energy storage SOC are set to 20% and 90% (the charging / discharging range of the energy storage system), the maximum discharge power is 100 MW, the maximum charging power is 90 MW (the power limit requirements of the load energy storage system), the frequency deviation shall not exceed ±0.3 Hz (ensuring that the power grid frequency fluctuation is controlled within an acceptable range), and the product of the single-time-step scheduling power and the tradable market volume shall not exceed 10 MWh (limiting the relationship between the power and volume of the energy storage participating in the market). The number of algorithm populations is 30, the maximum number of iterations is 100, the light intensity attenuation factor is 1, the perturbation parameter is 0.2, and the output results of the optimal strategy are as follows (selecting a part of them):

[0128]

[0129] Table 1 Data of the output results of the optimal strategy

[0130] After the scheduling is executed, the system monitors the response status, charge state, and frequency regulation revenue in real time. The actual monitoring results show that during the 7-day operation period, the cumulative frequency regulation revenue of the method of the present invention is 872,416 yuan, and the average daily revenue reaches 124,630 yuan, which is 19.2% higher than 732,209 yuan of the traditional static strategy; the average frequency regulation response delay is controlled within 0.6 seconds, and the system frequency deviation overrun rate is controlled within 1.2%, while the response delay of the traditional strategy is 1.5 seconds and the frequency overrun rate reaches 4.7%.

[0131] In addition, 10 groups of typical load disturbance scenarios are extracted from the training samples, and the scheduling response accuracy of the method of the present invention is compared with the baseline strategy. The following is the control response performance (unit: kW) of the load sudden increase scenario at 12:00 on a certain day:

[0132]

[0133] Table 2 Data of the control response performance of the load sudden increase scenario at 12:00 on a certain day

[0134] The present invention can achieve fast fitting of frequency regulation signals in each stage, and the average load response accuracy is improved by more than 18%, which can effectively reduce the risk of frequency anomalies caused by insufficient response.

[0135] In summary, the embodiments fully verify that the method of the present invention has good feasibility and significant economic optimization effects in the power trading scenario. By introducing a one-dimensional convolutional gated recurrent prediction model driven by revenue and a firefly optimization algorithm with outstanding global search ability, the system can not only accurately predict the future market and load status, but also achieve global optimization and real-time scheduling of strategies, solving the problems of lagging response, unstable revenue, and insufficient modeling of system constraints in traditional methods, providing effective support for the economic operation and intelligent control of independent energy storage systems.

[0136] The present invention introduces a multi-scale convolutional structure into the time series prediction model to extract the characteristics of power system changes at different time scales, combines the gated recurrent unit to model the long-term dependencies of electricity price, load, and state of charge, and simultaneously introduces a revenue weight vector to dynamically weight the importance of each feature in the frequency regulation revenue, guiding the network to focus on the key variables affecting economic benefits, realizing the deep coupling of the prediction output and the subsequent scheduling optimization goal. The structure is more target-oriented and can maintain high prediction stability and response ability under market conditions with strong volatility and frequent electricity price changes.

[0137] The present invention systematically integrates multiple actual operation constraints such as the boundary of the energy storage state of charge, charge and discharge power limits, energy balance equation, grid frequency deviation tolerance, and market trading capacity constraints in the optimization strategy modeling. The defined frequency regulation optimization strategy space not only maintains the global feasibility of the solution space, but also can fully express the coupling relationship between market dynamics and equipment operation, significantly improving the consistency between the optimization goal and engineering practice, and avoiding the model-system disconnection problem existing in existing methods.

[0138] The present invention combines the light intensity mechanism in the traditional firefly algorithm with the frequency regulation economic revenue function of the energy storage, encodes the predicted market electricity price and discharge strategy together as the fitness value of the objective function, and introduces a constraint verification and local perturbation mechanism in the individual update process, so that the algorithm is always controlled by the boundary constraints of the frequency regulation optimization strategy space during the search process, ensuring the legality of the solution and the executability of system operation. The firefly algorithm of the present invention has stronger dynamic adaptability and search accuracy, and realizes the global optimum and stable convergence of the scheduling strategy in the frequency regulation scenario with intertwined multiple factors.

[0139] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner can depend on the specific application and design constraint conditions of the technical solution. The technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the embodiments of the present disclosure.

[0140] The steps of the methods or algorithms described in connection with the embodiments disclosed herein may be implemented directly in hardware, in software modules executed by a processor, or in a combination thereof. The software modules may be disposed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0141] In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, apparatuses, etc.) may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units may be merely a logical function division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the couplings or direct couplings or communication connections shown or discussed with each other may be through some interfaces. The indirect couplings or communication connections of the devices or units may be in electrical, mechanical, or other forms.

[0142] Unless otherwise defined, the technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the technical field of the present invention. The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. Terms such as "portion" and "component" as used herein may represent either a single part or a combination of multiple parts. Terms such as "mounted" and "arranged" as used herein may represent either a component being directly attached to another component or a component being attached to another component through an intermediate member. Features described in one embodiment herein may be applied alone or in combination with other features to another embodiment, unless the feature is not applicable or otherwise stated in that other embodiment.

[0143] The present invention has been described by the above embodiments, but it should be understood that the above embodiments are only for the purpose of illustration and example, and are not intended to limit the present invention to the scope of the described embodiments. In addition, those skilled in the art can understand that the present invention is not limited to the above embodiments, and more variations and modifications can be made according to the teachings of the present invention, and these variations and modifications all fall within the scope claimed by the present invention. The protection scope of the present invention is defined by the appended claims and their equivalent scope.

Claims

1. A frequency modulation revenue optimization method for an independent energy storage system, characterized in that Including: S1. Obtain the power grid operation data set and the market transaction data set and standardize their formats; S2. Based on the standardized power grid operation data set and market transaction data set, construct a time series data set; S3. Use the time series data set to construct and train a one-dimensional convolutional gated recurrent network model to generate prediction results of future frequency regulation load and market electricity price; S4. According to the prediction results, construct a frequency regulation optimization problem aiming at maximizing the frequency regulation economic benefit, including the operation range of the energy storage system, the upper limit of the energy storage capacity, the system stability constraint and the market transaction rule constraint, and generate a frequency regulation optimization strategy space containing multiple independent energy storage frequency regulation strategies; S5. Through iterative search of the frequency regulation optimization strategy space, obtain the optimal independent energy storage frequency regulation strategy; S6. Control the energy storage device to perform frequency regulation operations according to the optimal independent energy storage frequency regulation strategy.

2. The method according to claim 1, wherein The power grid operation data in step S1 includes power grid frequency, power load, market electricity price and energy storage system status data, and the market transaction data includes the transaction volume of power transactions.

3. The method according to claim 1, wherein The step S2 includes: S21. Align the standardized power grid operation data set and market transaction data set according to a unified time step, and sequentially construct a multi-dimensional time series feature vector sequence sorted by time. The multi-dimensional time series feature vector sequence includes the power grid operation data and market transaction data at each time step; S22. In the multi-dimensional time series feature vector sequence, set a sliding time window with a fixed length, and divide the time series data into multiple consecutive sliding window subsequences in a sliding window manner. Each sliding window subsequence contains data of a fixed number of consecutive time steps, which is used to establish a model of the change process of the power system state in the time dimension, where the data dimension of each sliding window subsequence is equal to the product of the window length and the feature dimension; S23. Divide each sliding window subsequence into an input part and a target prediction part. The input part is composed of the feature vectors of the first several time steps of the sliding window subsequence and is used as the training input of the one-dimensional convolutional gated recurrent network model. The target prediction part is the feature vector of the last time step of the sliding window subsequence and is used as the training output of the one-dimensional convolutional gated recurrent network model, representing the future power grid operation state and market price behavior that the one-dimensional convolutional gated recurrent network model needs to predict, and construct a training sample set for all sliding window subsequences according to the division rules.

4. The method according to claim 3, wherein The training sample set is composed of sample pairs consisting of the input part and the corresponding target prediction part: Among them, represents the input part in the k-th sliding window subsequence, which is composed of multi-dimensional time series feature vectors of the first L w -1 time steps. The dimension of the multi-dimensional time series feature vector is 5, corresponding to the grid frequency, power load, market electricity price, state of charge of the energy storage system, and electricity trading volume. Y k represents the target prediction part in the k-th sliding window subsequence, and its dimension is the same as that of the feature vector at each time step in the input part of the sliding window subsequence T train represents the training sample set. k ∈ [1, T - L w + 1] represents the training sample index range. T is the total number of time steps, and L w is the sliding window length, that is, the number of time steps used for prediction within the sliding window.

5. The method according to claim 1, characterized in that, The step S4 includes: S41. Based on the trained one-dimensional convolutional gated recurrent network model, predict the power grid operation state and market price behavior in multiple future time steps, and construct a prediction input set. The prediction input set includes the predicted values of power grid frequency, power load, market electricity price, state of charge of the energy storage system and transaction volume of power transactions at each future time step; S42. Based on the market electricity price prediction value and the energy storage system scheduling discharge power in the predicted input set, construct the frequency modulation economic benefit function R within the future time range in combination with the capacity attenuation cost caused by the use of energy storage batteries profit : Among them, is the predicted value of the market electricity price, represents the dispatching discharge power of the energy storage system, is the capacity attenuation cost caused by the use of the energy storage battery; S43. Based on the predicted value of the state of charge of the energy storage system, the predicted value of the power grid frequency, the predicted value of the transaction volume of power transactions and the scheduled discharge power of the energy storage system in the prediction input set, establish a frequency regulation optimization constraint set; S44. Taking the FM economic benefit as the optimization objective and the FM optimization constraint set as the feasible solution space, construct the FM optimization strategy space function S opt : Among them, is the set of FM optimization constraints, represents the set of predicted inputs.

6. The method according to claim 5, characterized in that, The frequency modulation optimization constraint set includes: The state of charge boundary constraint of the energy storage system, which requires that the state of charge of the energy storage system at each future time step always remains between the preset minimum state of charge and the maximum state of charge; The power limit constraint, which requires that the energy storage power scheduling value at each time step does not exceed its maximum discharge power and maximum charge power; The energy balance constraint, which requires that the state of charge at each time step be updated according to the previous state of charge, battery efficiency, power value, and energy storage capacity; The frequency and market compliance constraint, which requires that the deviation of the grid frequency prediction value does not exceed the specified frequency tolerance range, and the product of the scheduling power of the energy storage system and the market trading volume at any time step does not exceed the maximum trading capacity limit allowed by the market.

7. The method according to claim 5, characterized in that, The step S5 includes: S51. Initialize the population of the firefly optimization algorithm based on the frequency modulation optimization strategy space function, where each firefly individual corresponds to an independent energy storage frequency modulation strategy P in the frequency modulation optimization strategy space k ; S52. For each independent energy storage frequency regulation strategy P k , calculate its frequency regulation economic benefit as the fitness function value according to the predicted input set : Among them, represents the discharge power part corresponding to the k-th independent energy storage frequency modulation strategy at the t-th time step, represents the capacity attenuation cost caused by the energy storage discharge behavior of the k-th independent energy storage frequency modulation strategy at the t-th time step; S53. Based on the fitness function value F k , according to the luminous intensity principle of the firefly optimization algorithm, define the luminous intensity I K of the k-th firefly individual as its frequency modulation economic benefit R profit (P k ), and calculate the mutual attraction β between firefly individuals according to the attraction update rule kj ; S54. Perform a position update operation on all independent energy storage frequency modulation strategies in each round of iteration, and update the frequency modulation strategy by combining the attraction mechanism and the local perturbation mechanism: Among them, is an independent energy storage frequency modulation strategy with higher fitness relative to the firefly individual k in the t-th round. ∈ is a local random perturbation factor, and rand() represents a random variable subject to a uniform distribution on [0, 1]; S55. After each position update, perform the following on the updated independent energy storage frequency regulation strategy for feasibility verification under the frequency regulation optimization constraint set including the state of charge boundary, power boundary, energy balance, frequency deviation, and market trading capacity upper limit constraint of the energy storage system. If any of the constraints are violated, roll back and correct the violation time step or randomly resample it to make each strategy firefly individual legal within the frequency regulation optimization strategy space; S56. Repeat steps S52 - S55 until the maximum number of iterations T is reached max or the global fitness value of the population meets the convergence criterion, and output the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy with the maximum frequency modulation benefit in the current population:

8. The method according to claim 7, wherein The step S6 includes: S61. Apply the energy storage power scheduling value P corresponding to the optimal independent energy storage frequency modulation strategy * to the power trading system to represent the energy storage power scheduling value of the optimal independent energy storage frequency modulation strategy at the t-th time step, where a positive value indicates discharging and a negative value indicates charging. The system sends the energy storage power scheduling value at each time step to the energy storage control module as a control instruction; S62. The energy storage system control unit receives and parses the control instruction at each time step, and drives the energy conversion module and the battery pack in the energy storage device to perform corresponding charge and discharge operations, so as to realize the hourly frequency modulation of the independent energy storage device. At the same time, the control instruction is connected to the power market management system to realize the collaborative response with the frequency modulation market instruction; S63. During the process of the energy storage device performing frequency modulation operation, multiple system operation state parameters are collected in real time to form an operation monitoring index set, including the actual response power, that is, the true output power of the energy storage system at the t-th time step, the state of charge of the energy storage system, that is, the SOC value at the t-th time step, the actual frequency modulation revenue obtained, that is, the real-time revenue calculated according to the market electricity price and the energy storage scheduling behavior, the grid frequency response value, that is, the grid frequency measured at the t-th time step, and the market transaction electricity price, that is, the electricity price at which the energy storage system participates in frequency modulation trading at the t-th time step; S64. Perform deviation analysis on the operation monitoring index set collected in real time, compare the execution effect of the optimal independent energy storage frequency modulation strategy with the actual operation state step by step in time, and calculate the deviation index; S65. If the cumulative result of the deviation index exceeds the set threshold, the system will judge that there is a performance decline or strategy deviation in the execution of the energy storage scheduling, and then trigger the adaptive feedback mechanism, and transmit the execution deviation information including power deviation, state of charge deviation, and revenue deviation back to the one-dimensional convolutional gated recurrent network model and the firefly optimization algorithm module; S66. After the frequency modulation strategy execution is completed, the frequency modulation execution data of the entire cycle is structurally integrated to generate a frequency modulation execution report.

9. The method according to claim 8, wherein The deviation index in the step S64 includes: Power deviation, that is, the absolute value of the difference between the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy and the actual corresponding power; State of charge deviation, that is, the difference between the actual state of charge of the energy storage system and the predicted state of charge; Revenue deviation, that is, the difference between the theoretical revenue calculated according to the actual market electricity price and the energy storage power scheduling value corresponding to the optimal independent energy storage frequency modulation strategy and the actual frequency modulation revenue.

10. The method according to claim 8 or 9, characterized in that, The frequency modulation execution report in step S66 includes the energy storage power scheduling value, actual power response value, state of charge change trajectory, grid frequency response, electricity market price, and final frequency modulation revenue data corresponding to the optimal independent energy storage frequency modulation strategy at each time step.

Images