A new energy primary frequency modulation optimization control method considering energy storage soc

Abstract

This invention discloses a primary frequency regulation optimization control method for new energy systems that considers the State of Energy (SOC) of energy storage. It overcomes the problem that traditional wind power control parameter tuning often relies on empirical formulas and fails to fully consider the dynamic changes in energy storage SOC. This method analyzes the power frequency characteristics of the grid-connected system and constructs a grid-connected frequency control model. Based on the high and low operating ranges of the energy storage SOC, it divides the system into correction coefficients and power allocation weights, calculates the total demand power based on frequency deviation, and allocates it to the wind turbine and energy storage system, establishing an adaptive frequency regulation power allocation strategy. A multi-objective function is constructed with the goals of minimizing frequency deviation, SOC fluctuation, and energy storage losses. An improved fuzzy neural network-deep reinforcement learning algorithm is used to identify operating conditions and iteratively correct control parameters, forming a closed-loop optimization to adapt to multiple operating conditions. This invention improves the frequency response speed and regulation margin of wind power under complex operating conditions while also considering energy storage lifespan, achieving synergistic optimization of frequency regulation performance and energy storage economy.

Description

Technical Field

[0001] This invention relates to a primary frequency regulation optimization control method for new energy sources that takes into account the State of Energy (SOC) of energy storage, and belongs to the field of power system frequency regulation control technology. Background Technology

[0002] Wind-storage frequency control refers to a technical means in power systems with a high proportion of wind power connected to the grid, where wind turbines and energy storage systems work together to provide the grid with more flexible and faster frequency regulation capabilities. This effectively alleviates the frequency fluctuation problems caused by the high proportion of renewable energy connected to the grid, improves the system's response to frequency disturbances, and ensures the safe and stable operation of the power grid. As traditional synchronous turbines are replaced by new energy sources, the proportion of wind power in the power system continues to increase, leading to a decline in system inertia and frequency regulation capabilities. Wind-storage synergy has become a key path to solve frequency stability problems.

[0003] Most existing wind-storage frequency control strategies use traditional empirical formulas to fix control parameters, failing to dynamically allocate frequency regulation power based on the real-time state of charge of energy storage and available power boundaries. Furthermore, they lack multi-objective collaborative optimization and adaptive closed-loop control mechanisms. These problems directly result in slow wind-storage frequency regulation response, low regulation accuracy, susceptibility to overcharging and over-discharging of energy storage, and poor adaptability under complex operating conditions.

[0004] Therefore, it is urgent to conduct in-depth research and design a new energy primary frequency regulation optimization control that takes into account energy storage SOC to improve the frequency response speed and regulation margin of wind power under complex operating conditions, and ensure the reliability and effectiveness of its frequency regulation service. Summary of the Invention

[0005] The purpose of this invention is to address the problems in existing wind-storage frequency control strategies, which mostly rely on traditional empirical formulas to fix control parameters, failing to dynamically allocate frequency regulation power based on the real-time state of charge (SOC) of energy storage and the available power boundary. Furthermore, these strategies lack multi-objective collaborative optimization and adaptive closed-loop control mechanisms. This invention proposes a new energy primary frequency regulation optimization control method that considers the SOC of energy storage. This method improves the response speed and adjustment margin of wind-storage joint frequency regulation, reduces energy storage losses, enhances system robustness, and fully taps the frequency regulation potential of wind farms. The core technical solutions are refined grid-connected frequency control modeling, adaptive frequency regulation power allocation based on the SOC of energy storage, and improved fuzzy neural network-deep reinforcement learning closed-loop parameter optimization. This approach focuses on solving the core problems of existing wind power participation in grid primary frequency regulation control strategies.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical means:

[0008] This invention proposes a primary frequency regulation optimization control method for new energy sources that considers the State of Charge (SOC) of energy storage, comprising the following steps:

[0009] The power frequency characteristics of the grid-connected system are analyzed, and a grid-connected frequency control model is constructed.

[0010] Based on the high and low ranges of energy storage SOC operation, correction coefficients and power allocation weights are divided, and a piecewise function of SOC correction coefficients is constructed. The total demand power of the grid primary frequency regulation is calculated from the frequency deviation and the cumulative effect of the frequency deviation. The total demand power is allocated to wind turbines and energy storage systems according to dynamic weights. The allocated power of wind turbines and energy storage systems after the first correction is obtained. SOC change rate correction term and response rate correction term are introduced to correct the allocated power of wind turbines and energy storage systems after the first correction. The allocated power of wind turbines and energy storage systems after the second correction is obtained, and a frequency regulation power adaptive allocation strategy is established.

[0011] A multi-objective function is constructed with the goals of minimizing frequency deviation, SOC fluctuation, and energy storage loss. An improved fuzzy neural network-deep reinforcement learning algorithm is used to identify operating conditions and iteratively correct control parameters.

[0012] As a further preferred embodiment of the present invention, a grid-connected system including wind power and energy storage is used as the analysis scenario. The power frequency characteristics of the grid-connected system are analyzed, and the expression of the grid-connected frequency control model is constructed as follows:

[0013]

[0014] In the formula, The system's overall inertial time constant; The overall damping coefficient of the system; This is for frequency deviation; This refers to the primary frequency regulation power deviation of the wind turbine unit. This refers to the primary frequency regulation power deviation of the energy storage system. This refers to the primary frequency regulation power deviation of conventional generating units; This refers to the power deviation of the power grid load.

[0015] As a further preferred embodiment of the present invention, the safe operating range of the energy storage SOC is set according to the high and low operating ranges of the energy storage SOC, and divided into three ranges: low, medium and high.

[0016] As a further preferred embodiment of the present invention, the expression for the piecewise function of the SOC correction coefficient is as follows:

[0017]

[0018] In the formula, This is the SOC correction factor for energy storage. This represents the lower limit of the correction factor for the low SOC region. This is the correction factor for the SOC region; This is the upper limit of the correction coefficient for the high SOC region; , Threshold for dividing the SOC interval, , These are the lower and upper limits of the safe operating range for energy storage SOCs.

[0019] The total power demand The calculation expression is as follows:

[0020]

[0021] In the formula, This represents the overall frequency droop factor of the system. For frequency deviation, The overall damping coefficient of the system. The cumulative time of frequency deviation. It is the integral of the power grid frequency deviation.

[0022] As a further preferred embodiment of the present invention, the power distribution of the wind turbine and energy storage system after one correction is obtained as follows:

[0023]

[0024] In the formula, This represents the power distribution of the wind turbine after a correction. This is the SOC correction factor for energy storage. For total power demand, The allocated power of the energy storage system after a correction. Assign weights to wind turbine power units; Assign weights to the power of the energy storage system;

[0025] The power allocation of the wind turbine and energy storage system after secondary correction is obtained as follows:

[0026]

[0027] in, This represents the power distribution of the wind turbine units after the second correction. The power allocation of the energy storage system after secondary correction. For response rate correction term, This is the correction term for the rate of change of SOC.

[0028] As a further preferred embodiment of the present invention, the expression of the multi-objective function is constructed with the objectives of minimizing frequency deviation, SOC fluctuation, and energy storage loss as follows:

[0029]

[0030] In the formula, For frequency deviation, For SOC fluctuations, For energy storage losses, , , These are the weighting coefficients;

[0031] The operating conditions are identified and control parameters are iteratively corrected by improving the fuzzy neural network-deep reinforcement learning algorithm, as follows:

[0032] The input variables are fuzzified to obtain the membership degrees of the fuzzy subsets of the input variables;

[0033] For each specific fuzzy rule, select the membership degree of the fuzzy category corresponding to each input variable under the fuzzy rule, and calculate the activation intensity under the fuzzy rule;

[0034] The activation intensity is deblurred, and the operating condition category is output.

[0035] Using the aforementioned working condition category as input and the control parameter correction amount as output, a deep reinforcement learning agent is constructed.

[0036] The extracted grid-connected system state feature vector and action feature vector are used as inputs, and the control parameter correction amount is based on the output of the deep reinforcement learning agent.

[0037] The control parameter correction is applied to the original control parameter to obtain the corrected control parameter.

[0038] As a further preferred embodiment of the present invention, the membership expression for obtaining the fuzzy subset of the input variable is as follows:

[0039]

[0040] In the formula, For the first The input variable belongs to the first The membership degree of a fuzzy subset. For input variables, As the center of the membership function, Width;

[0041] The expression for calculating the activation strength under the fuzzy rule is as follows:

[0042]

[0043] In the formula, For the first The activation strength of a fuzzy rule; The number of input variables; To select the membership degree of each input variable to a specific fuzzy category under each rule for that particular fuzzy rule, For the first Combination of fuzzy subsets of fuzzy rules;

[0044] The activation intensity is deblurred, and the output condition category is determined by the following formula:

[0045]

[0046] In the formula, Output values ​​for operating condition categories; For the first The output value of the fuzzy rule, This represents the total number of fuzzy rules.

[0047] As a further preferred embodiment of the present invention, the extracted grid-connected system state feature vector and action feature vector are used as input to establish a fully connected neural network as follows: ; In the formula, For parameters The network is in state ,action Below value, for The network can be trained with parameters; This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. For activation function, For state feature vectors, For action feature vectors, This involves concatenating vectors.

[0048] Target value The calculation expression is as follows:

[0049]

[0050] In the formula, Provide immediate rewards to intelligent agents; Discount factor; for The system state at any given moment. for Target action volume at any time. For the target network parameters, To select the maximum value at the next time step Value action.

[0051] As a further preferred embodiment of the present invention, a loss function is constructed to measure the prediction. Values ​​and Objectives The network parameters are iteratively updated based on the deviation of the values. The expression for the loss function is as follows:

[0052]

[0053] In the formula, To determine the loss value of the loss function, an adaptive learning rate is introduced to optimize the network parameters through iterative updates.

[0054] As a further preferred embodiment of the present invention, the expression for the corrected control parameters is as follows:

[0055]

[0056] In the formula, This is the corrected energy storage frequency regulation droop factor. The corrected energy storage response time constant, Assign weights to the corrected wind turbine power. The corrected SOC correction factor; This is the correction amount for the frequency regulation droop coefficient of energy storage. This is a correction factor for the energy storage response time constant. The correction amount for the power allocation weight of wind turbine units. This is the correction amount for the SOC correction factor. This is the SOC correction factor for energy storage. This refers to the droop factor for energy storage frequency regulation. The energy storage response time constant, Assign weights to wind turbine power units;

[0057] The corrected control parameters include , , as well as The corrected control parameters are fed back to the grid-connected frequency control model and the frequency regulation power adaptive allocation strategy in real time to complete the dynamic update of the control parameters.

[0058] Compared with existing technologies, the beneficial effects of this invention are as follows: Compared with traditional frequency regulation control strategies, the strategy proposed in this invention fully considers the volatility of wind power output and the time-varying nature of the grid-connected system's operating state. It constructs a multi-objective function with the goals of minimizing frequency deviation, SOC fluctuation, and energy storage loss. An improved fuzzy neural network-deep reinforcement learning algorithm is employed, combined with the real-time state of charge of energy storage and the available power boundary to dynamically allocate frequency regulation power. This dynamically tunes the wind power frequency regulation parameters and related control parameters. Furthermore, by combining the energy storage SOC adaptive power allocation strategy with real-time state updates, it can adaptively match the current frequency regulation needs of the power grid, achieving multi-objective collaborative optimization and an adaptive closed-loop control mechanism. While ensuring stable system operation and considering energy storage lifespan, it achieves synergistic optimization of frequency regulation performance and energy storage economy, significantly improving the frequency regulation accuracy, dynamic response speed, and robustness of the grid-connected system under complex operating conditions such as load disturbances and communication delays. Attached Figure Description

[0059] Figure 1 The flowchart below shows the new energy primary frequency regulation optimization control method considering the energy storage SOC of this invention. Figure 2 This is a schematic diagram of the grid connection frequency control model in an embodiment of the present invention; Figure 3 A comparison graph of the frequency response of different control strategies under load disturbance scenarios; Figure 4 This is a comparison curve of the frequency response of different control strategies in communication delay scenarios. Detailed Implementation

[0060] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0061] This invention proposes a primary frequency regulation optimization control method for new energy sources that considers the State of Charge (SOC) of energy storage, such as... Figure 1 As shown, the method of the present invention specifically includes the following steps:

[0062] Step A: Taking a grid-connected system containing wind power and energy storage as the analysis scenario, analyze the power frequency characteristics of the grid-connected system and construct a grid-connected frequency control model to provide a theoretical basis for subsequent analysis;

[0063] Step B: Divide the correction coefficient and power allocation weight according to the high and low range of the energy storage SOC operation, calculate the total demand power in combination with the frequency deviation and allocate it to the wind turbine and energy storage system, establish a frequency regulation power adaptive allocation strategy that takes into account the SOC change, and correct the allocation results of wind power and energy storage through the SOC change rate and frequency change rate to realize the dynamic control of the energy storage system.

[0064] Step C: Construct a multi-objective function with the goal of minimizing frequency deviation, SOC fluctuation and energy storage loss. Use an improved fuzzy neural network-deep reinforcement learning algorithm to identify operating conditions and iteratively correct control parameters to form a closed-loop optimization to adapt to multiple operating conditions.

[0065] In this embodiment of the invention, the specific operation of constructing the grid-connected frequency control module in step A is as follows:

[0066] Taking a system with wind power and energy storage connected to the grid as the analysis scenario, a refined grid-connected frequency control model was constructed based on the frequency response characteristics of wind power and energy storage equipment and the power frequency dynamic equation of the grid-connected system.

[0067] like Figure 2 As shown, the power frequency characteristics of the system are analyzed, and the expression for the grid-connected frequency control model is constructed as follows:

[0068]

[0069] In the formula, The system's overall inertial time constant; The overall damping coefficient of the system; This is for frequency deviation; This refers to the primary frequency regulation power deviation of the wind turbine unit. This refers to the primary frequency regulation power deviation of the energy storage system. This refers to the primary frequency regulation power deviation of conventional generating units; This refers to the power deviation of the power grid load (load power disturbance).

[0070] The primary frequency regulation power deviation of an AC power grid composed of conventional generating units is:

[0071]

[0072] In the formula, This refers to the droop coefficient of synchronous generator units in the power grid. , These are the turbine time constant and the power ratio coefficient of the high-pressure cylinder of the prime mover, respectively. Let be the Laplace operator, and let represent the complex frequency variable.

[0073] For wind turbine generators, considering the hysteresis characteristics of pitch angle adjustment and rotor speed dynamic response, the expression for its primary frequency regulation power deviation is:

[0074]

[0075] In the formula, This refers to the frequency regulation droop coefficient of the wind turbine. is the response time constant of the wind turbine.

[0076] For energy storage systems, considering SOC constraints and charge / discharge dynamic response characteristics, the expression for primary frequency regulation power deviation is as follows:

[0077]

[0078] In the formula, This is the SOC correction factor for energy storage; This refers to the droop factor for energy storage frequency regulation. The energy storage response time constant; The relationship with SOC will be quantified in subsequent steps.

[0079] In this embodiment of the invention, step B is performed as follows:

[0080] Based on the grid-connected frequency control model constructed above, real-time data collection is performed on energy storage SOC, frequency deviation, wind power output, and load. Wind power output data is reflected in the frequency response characteristics of wind turbines, indicating their output capacity and response lag, providing a basis for power allocation in frequency regulation. It is also reflected in adaptive frequency regulation power allocation, working in conjunction with energy storage output to dynamically adjust the power allocation ratio based on actual wind power output, ensuring precise matching between wind power output and frequency regulation requirements, while providing operating condition references for real-time correction of control parameters. Load data is reflected in the system power-frequency dynamic equation, serving as the grid load power deviation, reflecting actual load fluctuations, and is a crucial basis for measuring grid power demand and calculating total frequency regulation demand, directly affecting the frequency regulation power allocation logic. Therefore, by establishing an adaptive frequency regulation power allocation strategy considering SOC constraints, dynamic power coordination between wind power and energy storage is achieved. The specific steps and formulas are as follows:

[0081] Based on the high and low operating ranges of the energy storage SOC, the safe operating range of the energy storage SOC is set as follows: (Pick =20%, =80%), dividing the SOC into low, medium, and high ranges, constructing a piecewise function for the SOC correction coefficient, and realizing dynamic adjustment of frequency regulation capability under different SOC states, energy storage SOC correction coefficient The piecewise function expression is as follows:

[0082]

[0083] In the formula, This represents the lower limit of the correction factor for the low SOC region. This is the correction factor for the SOC region; This is the upper limit of the correction factor for the high SOC region. , These are the lower and upper limits of the safe operating range for energy storage SOCs. , The threshold values ​​for dividing the SOC interval are 40% and 60% respectively in this embodiment.

[0084] The total power demand for primary frequency regulation of the power grid is determined by both the frequency deviation and its cumulative effect. The total power demand is calculated accordingly. :

[0085]

[0086] In the formula, This represents the overall frequency droop factor of the system. It is the integral of the grid frequency deviation, reflecting the long-term cumulative demand of the grid frequency deviation on frequency regulation power. This represents the cumulative time of frequency deviation.

[0087] Based on the SOC state and total power demand, the total power demand... The power allocation formula is established by dynamically weighting the wind turbines and energy storage systems as follows:

[0088]

[0089] In the formula, This represents the power distribution of the wind turbine after a correction. The allocated power of the energy storage system after a correction. Assign weights to wind turbine power units; The power allocation weights of the energy storage system change dynamically with the SOC to achieve adaptive allocation of frequency regulation power.

[0090] Based on different SOC ranges, the power allocation weights of wind turbine units are further quantified. To achieve differentiated allocation of low, medium, and high SOC zones:

[0091]

[0092] In the formula, Because energy storage operates in a low SOC region, wind power undertakes more frequency regulation tasks; To ensure that energy storage is located in the mid-SOC zone, wind power and energy storage are evenly distributed; Because the energy storage is located in the high SOC region, it undertakes more frequency regulation tasks.

[0093] The power allocation weights for energy storage systems are derived from the power allocation weights for wind turbine units, ensuring the rationality of the allocation ratio.

[0094]

[0095] A SOC change rate correction term is introduced to prevent energy storage from having its lifespan damaged by overcharging and discharging. This term corrects the power allocation of the energy storage system. The expression for the SOC change rate correction term is as follows:

[0096]

[0097] In the formula, This is the correction factor for the rate of change of SOC; The real-time change rate of SOC is given, and the correction coefficient for the SOC change rate decreases as the SOC change rate increases.

[0098] A response rate correction term is introduced to prioritize the rapid adjustment capability of energy storage during periods of severe frequency fluctuations, thereby correcting the wind power distribution. The expression is as follows:

[0099]

[0100] In the formula, This is a response rate correction factor; The real-time rate of change of frequency is denoted as , and the response rate correction coefficient decreases as the rate of change of frequency increases.

[0101] Based on the SOC change rate correction term and the response rate correction term, after a second correction, the actual power allocation between the wind turbine and the energy storage is more suitable for the current operating conditions. The power allocation between the wind turbine and the energy storage system after the second correction is as follows:

[0102]

[0103] in, This represents the power distribution of the wind turbine units after the second correction. This refers to the power allocation of the energy storage system after the second correction.

[0104] In this embodiment of the invention, step C is performed as follows:

[0105] With the optimization objectives of minimizing frequency deviation, minimizing energy storage SOC fluctuation, and minimizing energy storage cycle loss, a multi-objective correction function is constructed to achieve synergistic optimization of frequency regulation performance and energy storage lifetime. The multi-objective correction function is expressed as follows:

[0106]

[0107] In the formula, This is the frequency deviation, i.e. The indicated power grid frequency deviation, For SOC fluctuations, For energy storage losses, , , These are weighting coefficients that can be adjusted according to grid demand. Each sub-objective function reflects frequency regulation accuracy, energy storage operation safety, and economy, respectively.

[0108] To quantify the contribution of frequency deviation to the optimization objective, a frequency deviation sub-objective function is constructed:

[0109]

[0110] In the formula, the integral term comprehensively reflects the cumulative degree of frequency deviation during frequency modulation; the smaller the deviation, the better. The lower the value.

[0111] To quantify the contribution of energy storage SOC fluctuations to the optimization objective, a sub-objective function for SOC fluctuations is constructed:

[0112]

[0113] In the formula, For energy storage batteries at all times State of charge at that time The smaller the integral term, the further the energy storage SOC deviates from the initial state, and the higher the operational safety.

[0114] To quantify the contribution of energy storage cycle losses to the optimization objective, a sub-objective function for energy storage losses is constructed:

[0115]

[0116] In the formula, For discrete time, This is the rated power of the energy storage. For discrete time The power distribution of energy storage after the second correction at any time, and the sum of the squares of the power proportions, can intuitively reflect the cycle loss of energy storage; the smaller the value, the lower the loss.

[0117] The improved FNN-DRL (Improved Fuzzy Neural Network-Deep Reinforcement Learning Algorithm) adopts a serial two-layer architecture. The first layer is the improved fuzzy neural network (FNN) operating condition recognition layer, which consists of a fuzzy input layer, a fuzzy rule layer, and a defuzzification output layer in sequence. It is used to fuzzify the real-time state variables of the power grid, perform rule activation calculations, and defuzzify the output to achieve accurate identification of operating conditions. The second layer is the deep reinforcement learning (DRL-DQN) parameter correction layer, which includes a state-action encoding layer, a fully connected Q-value network layer, and a parameter correction output layer. It takes the identified operating conditions and power grid state as inputs and outputs the core parameter correction quantities for primary frequency regulation through network mapping. The whole system takes the real-time state variables of the power grid as inputs and the parameter correction quantities as outputs, forming a closed-loop feedback optimization mechanism from operating condition identification to parameter optimization.

[0118] I. Improved Fuzzy Neural Network (FNN) (1) Fuzzy input layer

[0119] Fuzzification of input variables involves converting precise inputs into fuzzy membership degrees, representing the degree to which a variable belongs to each fuzzy subset. Utilizing the strong nonlinear mapping capability of an improved fuzzy neural network (FNN), multivariate operating condition features are extracted and their types identified, providing a basis for subsequent parameter correction. The fuzzification of input variables is first performed using the following formula:

[0120]

[0121] In the formula, For the first The input variable belongs to the first The membership degree of a fuzzy subset is used to fuzzify each input variable individually, reflecting the degree to which the input variable belongs to each of its own fuzzy categories, and is not bound to specific fuzzy rules. The input variables include frequency deviation, frequency change rate, SOC change rate, and total power demand, etc. As the center of the membership function, For width.

[0122] The parameters are set as follows: input dimension is 5; number of single-input fuzzy subsets is 3; center and width are determined by historical working condition clustering.

[0123] (2) Fuzzy rule layer

[0124] Based on the fuzzification results, for each specific fuzzy rule, the membership degree of the corresponding fuzzy category for each input variable under that rule is selected to calculate the activation strength of that fuzzy rule, thus achieving preliminary screening of the working condition features. The activation strength of each fuzzy rule is calculated as follows:

[0125]

[0126] In the formula, For the first The activation strength of a fuzzy rule; For the number of input variables, To select the membership degree of each input variable in a specific fuzzy category under that rule for each specific fuzzy rule; For the first The combination of fuzzy subsets of rules and the product operation can highlight the influence of core feature variables.

[0127] The parameters are set as follows: the total number of rules is 243.

[0128] (3) Deblurring output layer

[0129] The fuzzy activation intensity is mapped to continuous operating condition values, outputting a quantifiable operating condition identifier. The activation intensity is then defuzzified to output a clear operating condition category, thus completing the operating condition identification.

[0130]

[0131] In the formula, Output values ​​for operating condition categories; For the first The output value of the rule, The total number of fuzzy rules can be converted into clear operating condition identifiers by weighted summation.

[0132] The parameter settings are as follows: Output dimension is 1; Working condition category It can be discretized into operating conditions such as low SOC, high SOC, load disturbance, and communication delay.

[0133] II. Deep Reinforcement Learning (DRL-DQN) (1) State-Action Coding Layer By integrating the power grid status and FNN operating condition output, the DRL input status is constructed.

[0134]

[0135] In the formula, for The system state at any given moment has 5 dimensions. For frequency deviation, For the real-time rate of change of frequency, In the state of energy storage charge, For the real-time rate of change of SOC, Output operating condition values ​​for FNN. This is the transpose of a vector.

[0136] Define the DRL output action, which is the primary frequency modulation core parameter correction amount.

[0137]

[0138] In the formula, for Action vector at any moment, dimension 4; This is the correction amount for the frequency regulation droop coefficient of energy storage. This is a correction factor for the energy storage response time constant. The correction amount for the power allocation weight of wind turbine units. This is the correction amount for the SOC correction factor.

[0139] The original state and actions are mapped to high-dimensional feature vectors to enhance expressive power.

[0140]

[0141] In the formula, The state feature vector has a dimension of 64; The action feature vector has a dimension of 64. , For state mapping weights and biases, , Map weights and biases to actions.

[0142] (2) Fully connected Q-value network layer Using the working conditions identified by the FNN as input and the control parameter correction as output, a deep reinforcement learning agent (DQN) is constructed. This DQN serves as the core carrier of the deep reinforcement learning algorithm (DRL algorithm). Value network, computational objective The process of setting values ​​and iteratively updating network parameters enables real-time correction of the core control parameters of primary frequency modulation, which is the specific application of the DRL algorithm in this embodiment.

[0143] A Q-value network is established by interacting with the system to learn the optimal correction strategy. The extracted state feature vector and action feature vector of the grid-connected system are used as inputs to realize the value evaluation of state and action. The fully connected neural network is established as shown in the following formula:

[0144] ; In the formula, For parameters The network is in state ,action Below value, for The network can be trained with parameters; This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. For activation function, For state feature vectors, For action feature vectors, This involves concatenating vectors.

[0145] The parameters are set as follows: the network structure is a 2-layer fully connected network; the hidden layer dimension is 256; the activation function is ReLU; and the network parameters are initialized to a random normal distribution.

[0146] Calculation target This value provides optimization directions for parameter updates in deep reinforcement learning agents, improves the rationality of correction strategies, and aims to... value The calculation expression is as follows:

[0147]

[0148] In the formula, Provide immediate rewards to intelligent agents; Discount factor; for The system state at any given moment. for Target action volume at any time. For the target network parameters, To select the maximum value at the next time step Value action.

[0149] Frequency modulation performance, SOC stability, and energy storage loss are converted into rewards to construct a loss function that guides the optimization direction. Gradient descent is used to minimize the loss value, thereby optimizing the parameters of the deep reinforcement learning agent network. Iterative updates, loss value of the loss function The calculation expression is as follows:

[0150]

[0151] Expected squared error can effectively measure prediction Values ​​and Objectives The smaller the deviation of the value, the better the network parameters.

[0152] An adaptive learning rate is introduced to optimize the iterative updates of network parameters for deep reinforcement learning agents, while avoiding getting trapped in local optima. The calculation expression is as follows:

[0153]

[0154] In the formula, The initial learning rate, The learning rate decay period is defined as the learning rate decay cycle. Exponential decay allows the learning rate to gradually decrease as the number of iterations increases, balancing convergence speed and optimization accuracy.

[0155] The network parameters are iteratively updated to minimize the loss function and update the Q-value network parameters.

[0156]

[0157] In the formula, Update the parameters by assigning values. For the loss function with respect to parameters gradient, for The learning rate adapts to real-time conditions.

[0158] (3) Parameter correction output layer

[0159] The control parameter correction amount output by the deep reinforcement learning agent , , as well as Applying this to the original control parameters yields the corrected control parameters. , , as well as This enables dynamic optimization of control parameters; the corrected control parameter expressions are as follows:

[0160]

[0161] In the formula, For the corrected energy storage frequency regulation droop coefficient, The corrected energy storage response time constant, Assign weights to the corrected wind turbine power. The corrected SOC correction factor; This is the correction amount for the frequency regulation droop coefficient of energy storage. This is a correction factor for the energy storage response time constant. The correction amount for the power allocation weight of wind turbine units. This is the correction amount for the SOC correction factor.

[0162] The corrected control parameters are fed back in real time to the grid-connected frequency control model in step A and the frequency regulation power adaptive allocation strategy considering SOC constraints in step B, so as to complete the dynamic update of control parameters and form a closed-loop control of model construction-power allocation-parameter correction. This ensures that the system can take into account both frequency regulation performance and energy storage life under different energy storage SOC states and different grid frequency disturbance conditions, and achieve synergistic optimization of the two.

[0163] To verify the effectiveness of the present invention, the following example illustrates the proposed new energy primary frequency regulation optimization control method that considers energy storage SOC.

[0164] The constructed equivalent model was analyzed to demonstrate the trend of frequency fluctuations and verify the effectiveness of the optimization scheme proposed in this invention. A wind-storage control model was built in Matlab / Simulink, and the simulation system parameters are shown in Table 1.

[0165] Table 1. Simulation System Parameters

[0166]

[0167] To verify the effectiveness of the wind-storage frequency regulation parameter optimization control strategy proposed in this embodiment, simulations will be conducted using three different control scenarios, as detailed below:

[0168] Scenario 1: Using traditional frequency modulation control strategy;

[0169] Scenario 2: An adaptive frequency regulation power allocation strategy based on energy storage SOC is introduced, but the control parameters are not corrected in real time;

[0170] Scenario 3: Based on Scenario 2, the FNN-DRL algorithm is added to realize the real-time correction and closed-loop update of control parameters, which is the control strategy proposed in this invention.

[0171] Compare the frequency response characteristics of the system under load disturbance scenarios, such as Figure 3 As shown in the figure, the strategy proposed in this embodiment (Scenario 3) exhibits significant advantages in both frequency stability and dynamic response characteristics. The maximum frequency fluctuation value in Scenario 3 is only 0.348Hz, which is reduced by 25.8% and 11.5% compared to Scenario 1 and Scenario 2, respectively. This indicates that the strategy can effectively suppress frequency deviation caused by load disturbances and improve the frequency regulation accuracy of the system. Meanwhile, in terms of adjustment time, Scenario 3 only requires 6.09s to restore frequency stability, which is 30.4% and 20.2% shorter than Scenario 1 and Scenario 2, respectively, demonstrating faster dynamic response speed and system recovery capability. The comparative results show that the adaptive power allocation strategy of energy storage can dynamically adjust the power allocation weight of wind power and energy storage according to its SOC state, avoid overcharging and over-discharging of energy storage while optimizing frequency regulation response, and significantly improve the rationality of wind and energy storage coordinated frequency regulation. The FNN-DRL algorithm further enhances the adaptability of the control strategy to the randomness of wind power output and the uncertainty of load disturbance. It can accurately identify the operating conditions and correct the core control parameters in real time, making up for the shortcomings of fixed parameters that are difficult to adapt to load disturbance, and providing an effective solution for improving the frequency stability of new energy power systems.

[0172] To verify the frequency control performance under delayed scenarios and simulate the lag problem in command transmission in actual engineering, a 0.4s delay was set during command transmission. The control strategies under three scenarios were compared, and the system frequency response was as follows: Figure 4 As shown, the maximum frequency fluctuation in Scenario 3 is 0.374Hz, which is 35.1% and 20.9% lower than Scenario 1 and Scenario 2, respectively. This indicates that the strategy can effectively suppress frequency shift under delay conditions, demonstrating strong robustness and anti-interference ability. Furthermore, in terms of adjustment time, Scenario 3 only requires 8.56s to restore frequency stability, which is 41.9% and 17.9% shorter than Scenario 1 and Scenario 2, respectively, demonstrating faster dynamic response speed and system recovery capability. The comparative results show that the adaptive power allocation strategy based on energy storage SOC ensures the reasonable participation of energy storage in delay scenarios, while the FNN-DRL algorithm and closed-loop correction mechanism can identify delay conditions in real time and dynamically correct control parameters, offsetting the negative impact of delay and enhancing the adaptability of the control strategy to delay and system uncertainty.

[0173] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims.

Claims

1. A primary frequency regulation optimization control method for new energy sources considering energy storage SOC, characterized in that, Includes the following steps: The power frequency characteristics of the grid-connected system are analyzed, and a grid-connected frequency control model is constructed. Based on the high and low ranges of energy storage SOC operation, correction coefficients and power allocation weights are divided, and a piecewise function of SOC correction coefficients is constructed. The total demand power of the grid primary frequency regulation is calculated from the frequency deviation and the cumulative effect of the frequency deviation. The total demand power is allocated to wind turbines and energy storage systems according to dynamic weights. The allocated power of wind turbines and energy storage systems after the first correction is obtained. SOC change rate correction term and response rate correction term are introduced to correct the allocated power of wind turbines and energy storage systems after the first correction. The allocated power of wind turbines and energy storage systems after the second correction is obtained, and a frequency regulation power adaptive allocation strategy is established. A multi-objective function is constructed with the goals of minimizing frequency deviation, SOC fluctuation, and energy storage loss. An improved fuzzy neural network-deep reinforcement learning algorithm is used to identify operating conditions and iteratively correct control parameters.

2. The new energy primary frequency regulation optimization control method considering energy storage SOC as described in claim 1, characterized in that, Taking a grid-connected system containing wind power and energy storage as the analysis scenario, the power frequency characteristics of the grid-connected system are analyzed, and the expression of the grid-connected frequency control model is constructed as follows: ； In the formula, The system's overall inertial time constant; The overall damping coefficient of the system; This is for frequency deviation; This refers to the primary frequency regulation power deviation of the wind turbine unit. This refers to the primary frequency regulation power deviation of the energy storage system. This refers to the primary frequency regulation power deviation of conventional generating units; This refers to the power deviation of the power grid load.

3. The new energy primary frequency regulation optimization control method considering energy storage SOC as described in claim 1, characterized in that, Based on the high and low operating ranges of the energy storage SOC, the safe operating range of the energy storage SOC is set and divided into three ranges: low, medium, and high.

4. The new energy primary frequency regulation optimization control method considering energy storage SOC as described in claim 1, characterized in that, The expression for the piecewise function of the SOC correction coefficient is as follows: ； In the formula, This is the SOC correction factor for energy storage. This is the lower limit of the correction coefficient for the low SOC region; This is the correction factor for the SOC region; This is the upper limit of the correction coefficient for the high SOC region; , Threshold for dividing the SOC interval, , These are the lower and upper limits of the safe operating range for energy storage SOCs. The total power demand The calculation expression is as follows: ； In the formula, This represents the overall frequency droop factor of the system. For frequency deviation, The overall damping coefficient of the system. The cumulative time of frequency deviation. It is the integral of the power grid frequency deviation.

5. The new energy primary frequency regulation optimization control method considering energy storage SOC according to claim 1, characterized in that, The power allocation of the wind turbine and energy storage system after one correction is obtained as follows: ； In the formula, This represents the power distribution of the wind turbine after a correction. This is the SOC correction factor for energy storage. For total power demand, The allocated power of the energy storage system after a correction. Assign weights to wind turbine power units; Assign weights to the power of the energy storage system; The power allocation of the wind turbine and energy storage system after secondary correction is obtained as follows: ； in, This represents the power distribution of the wind turbine units after the second correction. The power allocation of the energy storage system after secondary correction. For response rate correction term, This is the correction term for the rate of change of SOC.

6. The new energy primary frequency regulation optimization control method considering energy storage SOC according to claim 1, characterized in that, The expression for the multi-objective function, which aims to minimize frequency deviation, SOC fluctuation, and energy storage loss, is as follows: ； In the formula, For frequency deviation, For SOC fluctuations, For energy storage losses, , , These are the weighting coefficients; The operating conditions are identified and control parameters are iteratively corrected by improving the fuzzy neural network-deep reinforcement learning algorithm, as follows: The input variables are fuzzified to obtain the membership degrees of the fuzzy subsets of the input variables; For each specific fuzzy rule, select the membership degree of the fuzzy category corresponding to each input variable under the fuzzy rule, and calculate the activation intensity under the fuzzy rule; The activation intensity is deblurred, and the operating condition category is output. Using the aforementioned working condition category as input and the control parameter correction amount as output, a deep reinforcement learning agent is constructed. The extracted grid-connected system state feature vector and action feature vector are used as inputs, and the control parameter correction amount is based on the output of the deep reinforcement learning agent. The control parameter correction is applied to the original control parameter to obtain the corrected control parameter.

7. The new energy primary frequency regulation optimization control method considering energy storage SOC according to claim 6, characterized in that, The membership expression for obtaining the fuzzy subset of the input variable is as follows: ； In the formula, For the first The input variable belongs to the first The membership degree of a fuzzy subset. For input variables, As the center of the membership function, Width; The expression for calculating the activation strength under the fuzzy rule is as follows: ； In the formula, For the first The activation strength of a fuzzy rule; The number of input variables; To select the membership degree of each input variable to a specific fuzzy category under each rule for that particular fuzzy rule, For the first Combination of fuzzy subsets of fuzzy rules; The activation intensity is deblurred, and the output condition category is determined by the following formula: ； In the formula, Output values ​​for operating condition categories; For the first The output value of the fuzzy rule, This represents the total number of fuzzy rules.

8. The new energy primary frequency regulation optimization control method considering energy storage SOC according to claim 6, characterized in that, Using the extracted state feature vector and action feature vector of the grid-connected system as input, a fully connected neural network is established as follows: ； In the formula, For parameters The network is in state ,action Below value, for The network can be trained with parameters; This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. For activation function, For state feature vectors, For action feature vectors, This involves concatenating vectors. Target value The calculation expression is as follows: ； In the formula, Provide immediate rewards to intelligent agents; Discount factor; for The system state at any given moment. for Target action volume at any time. For the target network parameters, To select the maximum value at the next time step Value action.

9. The new energy primary frequency regulation optimization control method considering energy storage SOC according to claim 8, characterized in that, Measure the prediction by constructing a loss function. Values ​​and Objectives The network parameters are iteratively updated based on the deviation of the values. The expression for the loss function is as follows: ； In the formula, To determine the loss value of the loss function, an adaptive learning rate is introduced to optimize the network parameters through iterative updates.

10. The new energy primary frequency regulation optimization control method considering energy storage SOC according to claim 6, characterized in that, The corrected control parameter expression is as follows: ； In the formula, This is the corrected energy storage frequency regulation droop coefficient. The corrected energy storage response time constant. Assign weights to the corrected wind turbine power. The corrected SOC correction factor; This is the correction amount for the frequency regulation droop coefficient of energy storage. This is a correction factor for the energy storage response time constant. The correction amount for the power allocation weight of wind turbine units. This is the correction amount for the SOC correction factor. This is the SOC correction factor for energy storage. This refers to the droop factor for energy storage frequency regulation. The energy storage response time constant, Assign weights to wind turbine power units; The corrected control parameters include , , as well as The corrected control parameters are fed back to the grid-connected frequency control model and the frequency regulation power adaptive allocation strategy in real time to complete the dynamic update of the control parameters.

Images