Self-adaptive smart power grid regulation and control method and system

Abstract

The invention provides a self-adaptive intelligent power grid regulation and control method and system, and belongs to the technical field of electric power. The method comprises the steps of firstly collecting historical operation data of a power grid, including power grid frequency fluctuation, a distributed energy output state and external disturbance information, then constructing a fractal chaotic attractor model, and deeply understanding a complex structure of dynamic behaviors of the power grid by calculating fractal dimensions and correlation coefficients of power grid state variables. Based on the model, the system can predict the stability boundary of the power grid, design fractal control signals of a self-similar structure, and inject the fractal control signals into the distributed energy control system to actively control the disturbance of the power grid. The operation state of the power grid and the external environment change are monitored in real time, fractal control parameters are dynamically adjusted through reinforcement learning, it is ensured that the control strategy is always matched with the actual working condition, and therefore the system performance is maximized. The anti-disturbance performance and the self-healing capability of the power grid are remarkably enhanced, and the energy utilization rate and the power supply quality are improved.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and in particular relates to an adaptive smart grid control method and system. Background Art

[0002] At present, smart grids are facing unprecedented challenges and opportunities. With the massive access of renewable energy and the popularization of distributed energy systems, the complexity of the power grid has increased significantly, making it difficult for traditional control strategies to dynamically, accurately and efficiently respond to real-time changes in the power grid. The turbulent fluctuations in the operation of the power grid, the uncertainty of the external environment, and internal faults and other variable factors all require the power grid control system to have a higher level of intelligence to ensure stability and self-healing capabilities. Traditional control methods are difficult to accurately predict power grid disturbances, and the control strategy is not proactive enough, resulting in slow response speed and inability to adapt to dynamic changes in the power grid, affecting the overall efficiency and safe operation stability of the power grid. Therefore, a more intelligent, dynamic and adaptable control method is needed to improve the comprehensive performance of the power grid. Summary of the invention

[0003] Based on the above technical problems, the present invention provides an adaptive smart grid control method and system, which realizes high-precision prediction and active control of grid disturbances by integrating the nonlinear dynamic analysis of chaos theory and the self-similar characteristics of fractal geometry, and significantly improves the anti-disturbance performance and self-healing ability of the grid.

[0004] The present invention provides an adaptive smart grid control method, the method comprising:

[0005] Step S1: Collect historical power grid operation data;

[0006] Step S2: construct a fractal chaos attractor model and calculate the fractal dimension of the power grid state variable;

[0007] Step S3: predicting the stability boundary of the power grid based on the fractal chaotic attractor model;

[0008] Step S4: designing a fractal control signal to generate a control signal with a self-similar structure;

[0009] Step S5: injecting the fractal control signal into the distributed energy control system;

[0010] Step S6: dynamically adjust the fractal control parameters according to the grid operation status and external environment changes.

[0011] Optionally, the constructing of a fractal chaotic attractor model and calculating the fractal dimension of a power grid state variable specifically includes:

[0012] Select key state variables and arrange the selected variables in chronological order to form a sequence;

[0013] Map the one-dimensional time series to a high-dimensional space and construct a state vector sequence, including:

[0014] Select embedding parameters, including embedding dimensions m and time delay τ ;

[0015] For each time point t, a state vector is constructed, expressed as:

[0016] X t =[f(t), f(t+τ),..., f(t+(m-1)τ), V(t), V(t+τ),..., V(t+(m-1)τ)]

[0017] For each time point t, based on the above m and τ, construct a state vector X containing the past m-1 time points and the current state t ; Through m and τ, the original one-dimensional time series is converted into a point set in the multidimensional state space, that is, each time point corresponds to an m-dimensional vector;

[0018] Calculate the correlation dimension D C , specifically including:

[0019] For any two state vectors X i , X j , calculate the distance between them, expressed as:

[0020]

[0021] In the formula, x ik and x jk They are the state vector X i and X j Elements in the kth dimension;

[0022] Count the number of neighborhood point pairs, set the neighborhood radius r, and count the state vector X of each point i The number of point pairs N in the neighborhood of i (r); that is, X i The number of state vectors in a sphere with a center and a radius of r;

[0023] Find the average number of neighborhood point pairs of all points and get N(r), which is expressed as:

[0024]

[0025] Where N is the total number of state vectors; N(r) describes how the average distribution density of points in the state space changes with radius;

[0026] Plot the correlation function and calculate the correlation dimension, expressed as:

[0027] Define the correlation function C(r) as the natural logarithm of the logarithm of the neighborhood points, that is, C(r) = ln[N(r)];

[0028] The slope D obtained by linear regression analysis C , expressed as:

[0029]

[0030] Where D C is the correlation dimension.

[0031] Optionally, the predicting the stability boundary of the power grid based on the fractal chaotic attractor model specifically includes:

[0032] The system is locally linearized and approximated as the Jacobian matrix J(X t , t), where X is the state vector and t is the time;

[0033] Perform iterative operations, including:

[0034] Initial vector setting, select a small initial vector difference δX0;

[0035] Iterative update, at each time step t, by δX t+1 = J(X t , t)·δX t Update vector difference;

[0036] Average growth rate calculation, calculate the average growth rate Λ over a period of time I , expressed as:

[0037]

[0038] In the formula, T is the length of time for analysis;

[0039] Repeating the iterative process for different orthogonal directions in the system state space, we obtain a series of Lyapunov exponents λ1, λ2, ..., λ n , forming a Lyapunov spectrum; if the maximum Lyapunov exponent λ max > 0, indicating that the system is in a chaotic state and has potential instability; if λ max <0, the system tends to be stable; by continuously tracking the maximum Lyapunov exponent λ max , find λ max = 0, identifying the critical point from stability to chaos, i.e., the stability boundary.

[0040] Optionally, the designing of a fractal control signal to generate a control signal having a self-similar structure specifically includes:

[0041] Using the fBm fractal model, Hurst exponent and time scale, the control signal is generated and expressed as:

[0042]

[0043] In the formula, B H (t) is the fBm value at time point t; t K =KΔt, Δt is the time interval; σ K is the standard deviation of each step based on the Hurst exponent adjustment; Z K is a variable with standard normal random distribution;

[0044] The generated control signal is fine-tuned by modifying its amplification factor or using a filter.

[0045] Optionally, dynamically adjusting the fractal control parameters according to the power grid operation state and external environment changes specifically includes:

[0046] Define reinforcement learning parameters, including:

[0047] Status t ∈S, represents the state of the power grid at time step t;

[0048] Action a t ∈A, the agent is in state s t The control action taken under the above conditions is to adjust the control parameters;

[0049] Reward t , the feedback the agent gets immediately after performing an action;

[0050] Maximizing the cumulative reward through the learning process, that is, finding the optimal strategy, is expressed as:

[0051]

[0052] Where π*(s) is the optimal strategy π* in a given state s; argmax π To find the strategy that maximizes the expression in brackets among all strategies π; is the expected value under strategy π, which is the average or expected sum of future rewards when strategy π is executed; t is the time step; r t is the immediate reward obtained at time step t; γ is the discount factor;

[0053] The implementation process includes:

[0054] Initialize parameters, set the initial strategy π0, define the state space S, action space A, reward function R(s, a) and discount factor γ;

[0055] At each time step t, the agent observes the current state of the grid s t ;

[0056] The agent follows the current policy π t , select action a t =π t (s t ), and apply it to the power grid, observing the next state s t+1 and instant rewards t ;

[0057] Use Q-learning algorithm according to s t , a t , r t ,s t+1 Update strategy π t , optimize long-term rewards;

[0058] The learned policy is directly converted into a dynamic adjustment of the fractal control parameters.

[0059] The present invention also provides an adaptive smart grid control system, the system comprising:

[0060] Data collection module, used to collect historical operation data of power grid;

[0061] A model building module is used to build a fractal chaos attractor model and calculate the fractal dimension of the power grid state variables;

[0062] A boundary prediction module, which predicts the stability boundary of the power grid based on the fractal chaotic attractor model;

[0063] Signal generation module, designs fractal control signals and generates control signals with self-similar structures;

[0064] A system control module for injecting fractal control signals into a distributed energy control system;

[0065] The parameter optimization module dynamically adjusts the fractal control parameters according to the grid operation status and external environment changes.

[0066] Optionally, the model building module specifically includes:

[0067] Select key state variables and arrange the selected variables in chronological order to form a sequence;

[0068] Map the one-dimensional time series to a high-dimensional space and construct a state vector sequence, including:

[0069] Select embedding parameters including embedding dimension m and time delay τ ;

[0070] For each time point t, a state vector is constructed, expressed as:

[0071] X t =[f(t), f(t+τ),..., f(t+(m-1)τ), V(t), V(t+τ),..., V(t+(m-1)τ)]

[0072] For each time point t, based on the above m and τ, construct a state vector X containing the past m-1 time points and the current state t ; Through m and τ, the original one-dimensional time series is converted into a point set in the multidimensional state space, that is, each time point corresponds to an m-dimensional vector;

[0073] Calculate the correlation dimension D C , specifically including:

[0074] For any two state vectors X i , X j , calculate the distance between them, expressed as:

[0075]

[0076] In the formula, x ik and x jk They are the state vector X i and X j Elements in the kth dimension;

[0077] Count the number of neighborhood point pairs, set the neighborhood radius r, and count the state vector X of each point i The number of point pairs N in the neighborhood of i (r); that is, X i The number of state vectors in a sphere with a center and a radius of r;

[0078] Find the average number of neighborhood point pairs of all points and get N(r), which is expressed as:

[0079]

[0080] Where N is the total number of state vectors; N(r) describes how the average distribution density of points in the state space changes with radius;

[0081] Plot the correlation function and calculate the correlation dimension, expressed as:

[0082] Define the correlation function C(r) as the natural logarithm of the logarithm of the neighborhood points, that is, C(r) = ln[N(r)];

[0083] The slope D obtained by linear regression analysis C , expressed as:

[0084]

[0085] Where D C is the correlation dimension.

[0086] Optionally, the boundary prediction module specifically includes:

[0087] The system is locally linearized and approximated as the Jacobian matrix J(X t , t), where X is the state vector and t is the time;

[0088] Perform iterative operations, including:

[0089] Initial vector setting, select a small initial vector difference δX0;

[0090] Iterative update, at each time step t, by δX t+1 = J(X t , t)·δX t Update vector difference;

[0091] Average growth rate calculation, calculate the average growth rate Λ over a period of time I , expressed as:

[0092]

[0093] In the formula, T is the length of time for analysis;

[0094] Repeating the iterative process for different orthogonal directions in the system state space, we obtain a series of Lyapunov exponents λ1, λ2, ..., λ n , forming a Lyapunov spectrum; if the maximum Lyapunov exponent λ mmax > 0, indicating that the system is in a chaotic state and has potential instability; if λ max <0, the system tends to be stable; by continuously tracking the maximum Lyapunov exponent λ max , find λ max = 0, identifying the critical point from stability to chaos, i.e., the stability boundary.

[0095] Optionally, the signal generating module specifically includes:

[0096] Using the fBm fractal model, Hurst exponent and time scale, the control signal is generated and expressed as:

[0097]

[0098] In the formula, B H (t) is the fBm value at time point t; t K=KΔt, Δt is the time interval; σ K is the standard deviation of each step based on the Hurst exponent adjustment; Z K is a variable with standard normal random distribution;

[0099] The generated control signal is fine-tuned by modifying its amplification factor or using a filter.

[0100] Optionally, the parameter optimization module specifically includes:

[0101] Define reinforcement learning parameters, including:

[0102] Status t ∈S, represents the state of the power grid at time step t;

[0103] Action a t ∈A, the agent is in state s t The control action taken under the above conditions is to adjust the control parameters;

[0104] Reward t , the feedback the agent gets immediately after performing an action;

[0105] Maximizing the cumulative reward through the learning process, that is, finding the optimal strategy, is expressed as:

[0106]

[0107] Where π*(s) is the optimal strategy π* in a given state s; argmax π To find the strategy that maximizes the expression in brackets among all strategies π; is the expected value under strategy π, which is the average or expected sum of future rewards when strategy π is executed; t is the time step; r t is the immediate reward obtained at time step t; γ is the discount factor;

[0108] The implementation process includes:

[0109] Initialize parameters, set initial strategy π0, define state space S, action space A, reward function R(s, a) and discount factor y;

[0110] At each time step t, the agent observes the current state of the grid s t ;

[0111] The agent follows the current policy π t , select action a t =π t (s t ), and apply it to the power grid, observing the next state s t+1 and instant rewardst ;

[0112] Use Q-learning algorithm according to s t , a t , r t ,s t+1 Update strategy π t , optimize long-term rewards;

[0113] The learned policy is directly converted into a dynamic adjustment of the fractal control parameters.

[0114] Compared with the prior art, the present invention has the following beneficial effects:

[0115] The present invention realizes the transformation from passive response to active control through high-precision prediction of power grid disturbances, and can take measures to prevent accidents before the stability of the power grid is threatened; it integrates chaos theory and fractal geometry characteristics to enable the power grid to show higher stability and resilience when facing internal and external disturbances; through adaptive adjustment of control strategies, the power grid can quickly return to normal operating conditions after encountering faults and achieve self-repair; it uses reinforcement learning to automatically optimize fractal control parameters, realizes the intelligence and dynamic nature of the scheduling strategy, and improves the operating efficiency of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] Figure 1 A flow chart of an adaptive smart grid control method of the present invention;

[0117] Figure 2 This is a structural diagram of an adaptive smart grid control system of the present invention. DETAILED DESCRIPTION

[0118] The present invention is further described below in conjunction with specific implementation cases and drawings, but the present invention is not limited to these embodiments.

[0119] Example 1

[0120] like Figure 1 As shown, the present invention discloses an adaptive smart grid control method, the method comprising:

[0121] Step S1: Collect historical power grid operation data.

[0122] Step S2: construct a fractal chaos attractor model and calculate the fractal dimension of the power grid state variables.

[0123] Step S3: predicting the stability boundary of the power grid based on the fractal chaotic attractor model.

[0124] Step S4: Design a fractal control signal to generate a control signal with a self-similar structure.

[0125] Step S5: injecting the fractal control signal into the distributed energy control system.

[0126] Step S6: dynamically adjust the fractal control parameters according to the grid operation status and external environment changes.

[0127] The following is a detailed discussion of each step:

[0128] Step S1: Collect historical grid operation data, including but not limited to grid frequency fluctuation records, distributed energy output status and external disturbance information.

[0129] Step S1 specifically includes:

[0130] Data types, including:

[0131] Grid frequency fluctuation records. Grid frequency (usually refers to the frequency of alternating current, such as 50Hz or 60Hz) is a key indicator for measuring the stability of the power system. Frequency fluctuation records include time series data, showing the changes in grid frequency over different time periods. These changes may be caused by factors such as supply and demand imbalances, large load mutations, and generator failures. Recording these fluctuations helps analyze the response characteristics of the grid, detect abnormal events, and evaluate the system's regulation capabilities.

[0132] Distributed energy output status. With the development of renewable energy and distributed power generation technology, it is particularly important to record the output power and start and stop status of each distributed energy source (such as solar photovoltaic, wind power, micro hydropower station, etc.); it not only reflects the contribution of distributed energy to the power grid, but also is an important basis for evaluating the flexibility of the power grid, planning and scheduling, and balancing supply and demand.

[0133] External disturbance information: Grid operation is affected by many external factors, such as weather conditions (affecting renewable energy output), large-scale user behavior patterns (such as peak electricity consumption during holidays), natural disasters (causing damage to transmission lines), sudden load changes (such as the start-up or shutdown of large factories), grid fault records, and any external events that may affect grid operation; it helps to predict possible grid stress situations and take countermeasures in advance.

[0134] Data sources include:

[0135] Sensor network, sensors deployed at key nodes of the power grid (such as voltage, current, and frequency sensors) continuously monitor the status of the power grid and transmit data to the data center or cloud platform in real time.

[0136] Smart meters installed at the user end can collect detailed electricity usage information and reflect the user's electricity usage pattern, which is crucial for demand-side management.

[0137] Weather stations and environmental monitoring systems are used to collect environmental data that affect distributed energy output, such as wind speed, light intensity, temperature, etc.

[0138] SCADA system (Supervisory Control and Data Acquisition System) is widely used in modern power grids to centrally monitor and control power grid equipment and collect a large amount of real-time operating data.

[0139] Data collection and integration, including:

[0140] Timed data collection: set the data collection frequency to ensure the continuity and integrity of the data; for key parameters (such as grid frequency), collection at time intervals of seconds or shorter may be required.

[0141] The collected raw data needs to be preprocessed (such as cleaning, verification, and standardization) for analysis.

[0142] Data cleaning to remove invalid, erroneous or abnormal data records, such as abnormal readings caused by sensor failure.

[0143] Standardization and formatting: converting data from different sources and formats into a standard format suitable for analysis to facilitate subsequent processing.

[0144] Data applications include:

[0145] Historical data analysis: By analyzing historical data, we can discover the operating rules of the power grid, identify potential risk points, and optimize operating strategies.

[0146] Model building and prediction serve as the basis for constructing a power grid model (such as the fractal chaos attractor model in step S2) to simulate and predict power grid behavior.

[0147] Control strategy formulation: Based on data insights, formulate more accurate control strategies and dispatch plans to improve the stability and efficiency of the power grid.

[0148] Step S2: construct a fractal chaos attractor model and calculate the fractal dimension and correlation coefficient of the power grid state variables.

[0149] Step S2 specifically includes:

[0150] Data preprocessing, including:

[0151] Ensure that the historical grid operation data (frequency fluctuations, energy output, external disturbances, etc.) obtained from S1 is clean and continuous; remove outliers through data cleaning and reduce random noise through smoothing to ensure data quality; this means converting all measured grid variables to the same scale, such as by standardization (subtracting the mean and dividing by the standard deviation) or normalization (adjusting to a specific interval such as 0 to 1); the purpose is to eliminate the differences in magnitude between different variables so that comparisons are meaningful.

[0152] Select key state variables, such as grid frequency f(t) and key node voltage V(t), to construct time series; the operating state of the power grid can be described by multiple variables, but not all of them are equally important for analyzing chaotic characteristics; selecting variables such as grid frequency f(t) and key node voltage V(t) as representatives is directly related to the stability and efficiency of the power grid.

[0153] Construct a time series and arrange the selected variables in chronological order to form a sequence, which provides basic data for subsequent dynamic analysis.

[0154] Map the one-dimensional time series to a high-dimensional space and construct a state vector sequence, including:

[0155] Select embedding parameters including embedding dimension m and time delay τ .

[0156] The embedding dimension m determines how many continuous states (for example, the values ​​of grid frequency and voltage at different time points) are used to construct a state vector; it is usually determined using the false nearest neighbor criterion to ensure that the topological structure between data points in high-dimensional space can be correctly reflected.

[0157] The time delay τ refers to the time interval between adjacent state variables when constructing the state vector. The first minimum point that decays to zero is found by calculating the autocorrelation function, and this is used as the estimated value of τ to ensure that the state variables are correlated but not excessively overlapped.

[0158] In this embodiment, it is determined how many consecutive time points are needed to describe the state of the system (for example, information from the past four time points is selected to predict the state of the fifth point, and the embedding dimension m is 4 at this time), as well as the time interval between these points (time delay τ).

[0159] For each time point t, a state vector is constructed, expressed as:

[0160] X t =[f(t), f(t+τ),..., f(t+(m-1)τ), V(t), V(t+τ),..., V(t+(m-1)τ)]

[0161] For each time point t, based on the above m and τ, construct a state vector X containing the past m-1 time points and the current state t , to reflect the dynamic behavior of the system; based on the selected parameters, a vector containing multiple variables is constructed for each time point; if the grid frequency and voltage are considered, and m=2, τ=1, then a state vector may be [f(t), V(t+1)].

[0162] Through m and τ, the original one-dimensional time series is converted into a set of points in the multidimensional state space, that is, each time point corresponds to an m-dimensional vector.

[0163] Calculate the correlation dimension D C , specifically including:

[0164] For any two state vectors X i , X j , calculate the distance between them, expressed as:

[0165]

[0166] In the formula, x ik and x jk They are the state vector X t and X j For elements in the kth dimension, the distance measures the proximity of two points in the state space and is the basis for subsequent calculations.

[0167] Count the number of neighborhood point pairs, set the neighborhood radius r, and count the state vector X of each point i In the neighborhood of ij The number of point pairs N that are smaller than all points of r i (r); that is, X i The number of state vectors within a sphere with a center and a radius of r.

[0168] Find the average number of neighborhood point pairs of all points and get N(r), which is expressed as:

[0169]

[0170] Where N is the total number of state vectors; N(r) describes how the average distribution density of points in the state space changes with radius.

[0171] Plot the correlation function and calculate the correlation dimension, expressed as:

[0172] The correlation function C(r) is defined as the natural logarithm of the logarithm of the neighborhood points, that is, C(r) = ln[N(r)].

[0173] The relationship between C(r) and ln[r] is depicted through a scatter plot. If the system exhibits a certain self-similarity, when r is small, C(r) and ln[r] show a linear relationship.

[0174] If there is a linear region, the slope D obtained by linear regression analysis C , which is the correlation dimension, expressed as:

[0175]

[0176] In this embodiment, the selection of the neighborhood radius r, if the points are very dense, a smaller r may be more appropriate because it can more finely distinguish the points in the neighborhood; conversely, if the points are sparse, a larger r may be required to ensure that each neighborhood has enough points for statistical analysis; if the system exhibits self-similarity, then the structures at different scales should be similar. When selecting r, it is necessary to ensure that it can span at least several self-similar intervals so that a clear linear region can be observed on the correlation function graph, thereby accurately estimating the correlation dimension; in a limited data set, an r that is too large will lead to a boundary effect, that is, when the state vector is close to the boundary of the data set, its neighborhood is truncated and cannot truly reflect the global characteristics.

[0177] In this embodiment, the method for calculating the fractal dimension adopts the correlation dimension method or the maximum Lyapunov exponent estimation, because they are suitable for the analysis of chaotic systems.

[0178] For two state vectors X i , X j , calculate its correlation coefficient C ij , understanding the interactions and dependencies between various state variables helps to identify the coupling mechanisms within the system.

[0179] In this embodiment, the above content obtains a point set in a high-dimensional state space (representing the dynamic behavior of the power grid), and also understands the complex structure of these point sets through fractal dimension, reveals the unpredictability of the system through chaotic characteristic analysis, and explores the interaction within the system through the correlation coefficient matrix. All of these combined together constitute a "fractal chaos attractor model", which is not only a high-level abstraction of the dynamic behavior of the power grid, but also a theoretical basis for predicting future states and formulating control strategies.

[0180] Step S3: predicting the stability boundary of the power grid based on the fractal chaotic attractor model.

[0181] Step S3 specifically includes:

[0182] Starting from the state space reconstruction results obtained from S2, a sufficiently long stable state sequence is selected to ensure the reliability of the analysis; the state space reconstruction obtains a multidimensional representation of the dynamic behavior of the power grid, namely a series of m-dimensional state vectors.

[0183] The system is locally linearized and approximated as the Jacobian matrix J(X t , t), where X is the state vector and t is the time; the Jacobian matrix describes the small-scale linear behavior of the system at that point and is the basis for calculating the Lyapunov exponent.

[0184] Perform iterative operations, including:

[0185] Initial vector setting, select a small initial vector difference δX0, usually a unit vector, as the starting point of the divergence measure.

[0186] Iterative update, at each time step t, by δX t+1 = J(X t , t)·δX t Update the vector difference to simulate the divergence or convergence process of adjacent orbits.

[0187] Average growth rate calculation, calculate the average growth rate Λ over a period of time I , expressed as:

[0188]

[0189] Where T is the length of time for the analysis. This step is obtained by calculating the natural logarithm of the growth rate of the final vector difference compared to the initial value and taking the average.

[0190] Repeat the above iterative process for different orthogonal directions in the system state space, and finally obtain a series of Lyapunov exponents λ1, λ2, ..., λ n , constituting the Lyapunov spectrum; the maximum Lyapunov exponent is particularly critical because it is directly related to the chaotic properties of the system.

[0191] If the maximum Lyapunov exponent λ max > 0, indicating that the system is in a chaotic state and has potential instability; if λ max <0, the system tends to be stable.

[0192] By continuously tracking the maximum Lyapunov exponent λ max As system parameters or external conditions change, find λ max = 0, which is the boundary where the system turns from stability to instability. The critical point from stability to chaos is identified, that is, the stability boundary.

[0193] Predicting instability trends: If we observe that λ in some regions max Gradually increasing over time, this may indicate that the system is approaching instability; by analyzing these trends, steps can be taken to prevent potential failures in advance.

[0194] In actual power grid operation, the maximum Lyapunov exponent is continuously monitored. Once it approaches or exceeds the threshold, an early warning signal is issued, indicating that the system may be about to enter an unstable state. Based on the early warning information, preventive measures are taken, such as adjusting the output of the generator set, changing the load distribution and other control strategies, to slow down or prevent the system from entering an unstable state. Lyapunov exponent analysis not only provides early warnings, but also guides how to optimize system performance and stability through parameter adjustment.

[0195] In this embodiment, the Lyapunov index is an indicator of the divergence or convergence rate of adjacent orbits in the system state space, and is a core tool for chaotic system analysis. For an n-dimensional dynamic system, there are n Lyapunov exponents. The positive value of the maximum Lyapunov exponent indicates that the system has chaotic behavior, and a negative value means that the system tends to a stable fixed point or periodic motion.

[0196] Step S4: Design a fractal control signal to generate a control signal with a self-similar structure.

[0197] Step S4 specifically includes:

[0198] Fractal theory is a discipline that studies non-integer dimensional geometric forms that are ubiquitous in nature. It reveals characteristics such as self-similarity and scale invariance in complex systems. In power systems, phenomena such as load changes and voltage fluctuations in power grids often exhibit fractal characteristics, which provides a theoretical basis for designing control signals that match the natural fluctuation patterns of the power grid.

[0199] Common fractal generation models include fractal Brownian motion fB m , Hurst exponential model, Mandelbrot set, etc.; In this scenario, fBm is particularly suitable for generating control signals that match the power grid dynamics because it can well simulate random processes with long-term memory; fB m The mathematical definition of can be characterized by the Hurst index H, where H∈(0, 1) corresponds to different types of self-similarity.

[0200] Based on the fractal characteristic analysis of the power grid in step S2, the Hurst index of the power grid fluctuation is determined to reflect the long-term memory or anti-memory of the signal, and the fractal characteristics of the data sequence in the power grid are identified and quantified, which involves statistical analysis of the historical monitoring data to capture its self-similarity and scale invariance characteristics.

[0201] Statistical methods such as R / S analysis (rescaled range analysis) and Hurst index calculation method are used to estimate the Hurst index of power grid dynamics; the Hurst index value ranges from 0 to 1, H < 0.5 indicates that the signal has anti-persistence or negative correlation, H = 0.5 indicates random walk, and H > 0.5 means that the signal has long-term memory or positive correlation.

[0202] According to the actual frequency of grid operation and the instability trend identified in step S3, the time scale of the control signal is determined to ensure that the signal can effectively act on the key dynamic process of the system.

[0203] By analyzing the real-time monitoring data and historical data of the power grid, any trends that may indicate system instability can be identified, including changes in indicators such as increased power fluctuations, abnormal frequency, and increased voltage fluctuations. Time series analysis, machine learning algorithms, or signal processing techniques can be used to detect abnormal patterns and periodic changes in the data to predict future instability.

[0204] Based on the instability trends identified by S3, for example, if it is found that certain instability events tend to occur within a specific time period (such as daily cycles, weekly cycles or seasonal changes), the time scale of the control signal should match it; this means that the design of the control signal should take into account these critical time windows to ensure that the frequency and duration of the signal can respond to and intervene in these instability trends in a timely manner, so as to maximize the effectiveness when the system needs control intervention the most.

[0205] Using the selected fractal model (fBm), combined with the determined Hurst exponent and time scale, the control signal is generated and expressed as:

[0206]

[0207] In the formula, B H (t) is the fBm value at time point t; t K =KΔt, Δt is the time interval; σ K is the standard deviation of each step based on the Hurst exponent adjustment; Z K is a standard normal random distributed variable.

[0208] This process is implemented using a computer program (such as the fractionalBrownianMotion function of MATLAB) to generate a signal sequence with specified self-similar properties by inputting the Hurst exponent H and the required time series length.

[0209] By modifying the signal's amplification factor or using a filter to fine-tune the signal, the signal can be more consistent with the details of the power grid fluctuations; by adjusting the local intensity or frequency component of the signal, the generated control signal not only matches the power grid fluctuations in terms of statistical characteristics, but also presents a self-similar structure in waveform that is similar to the natural fluctuations of the power grid.

[0210] Step S5: Inject the fractal control signal into the distributed energy control system, and realize multi-scale fractal synchronization between different nodes through communication protocols and algorithms.

[0211] Step S5 specifically includes:

[0212] First, the distributed energy control system can be regarded as composed of multiple subsystems or nodes, each of which is responsible for monitoring and controlling the energy distribution and consumption in a specific area. Assume that the system contains Num nodes, and the dynamic behavior of each node can be described by a nonlinear dynamic equation, which includes chaotic characteristics. Taking a typical chaotic Lorenz system as an example, the dynamic equation of the Jth node can be expressed as:

[0213]

[0214] In the formula, x J ,y J , z J is the state variable of node J, σ, ρ are system parameters; u J (t) is the fractal control signal optimized in step S4, which aims to guide the chaotic behavior of each node into a synchronous state, and the signal must have the characteristic of being self-similar to the natural fluctuation pattern of the system.

[0215] In order to achieve multi-scale fractal synchronization, a master-slave synchronization or mutual synchronization strategy is adopted, and combined with distributed control theory, this application adopts a differential synchronization method, in which each node attempts to reduce the difference in state with adjacent nodes; assuming that the nodes are connected through a network, information exchange follows a specific communication protocol, such as Zigbee or a customized wireless sensor network protocol, to achieve real-time sharing of state information, the differential synchronization equation is expressed as:

[0216] Δx J =q(x J+1 -x J ), J = 1, ..., Num-1

[0217] In the formula, Δx J is the state difference between nodes J and J+1, and q is the synchronization gain; by adjusting q, the system can be made to reach a global synchronization state.

[0218] Distributed fractal chaos synchronization not only improves the system response speed, but also significantly enhances its anti-disturbance ability. When the system suffers from external interference or internal failure, the synchronization mechanism prompts each node to quickly correct each other, and achieves rapid recovery of the global state through local interaction, reflecting the characteristics of self-repair. This is because the introduction of fractal control signals enhances the nonlinear dynamic stability of the system, allowing the system to maintain or quickly recover to the synchronized state when facing disturbances, effectively preventing the propagation and amplification of disturbances.

[0219] In this embodiment, the above steps implement distributed fractal chaotic synchronization control by integrating the optimized fractal control signal into each node of the distributed energy control system, and using specific communication protocols and algorithms to achieve real-time status information exchange and synchronization, thereby improving the overall anti-disturbance ability and self-repair function of the system; it makes full use of the complex dynamic characteristics of the chaotic system and the self-similarity of the fractal signal at multiple scales, providing a new and efficient control method for the field of distributed energy control.

[0220] Step S6: According to the current operation state of the power grid and changes in the external environment, dynamically adjust the fractal control parameters, optimize the control strategy, and achieve adaptive optimization of the control parameters.

[0221] Step S6 specifically includes:

[0222] Define reinforcement learning parameters, including:

[0223] Environment, the power grid and its external environment, including but not limited to load changes, renewable energy fluctuations, equipment failures, etc.; the state of the power grid can be represented by a series of observable variables, such as the voltage, current, frequency, etc. of each node.

[0224] Agent, the control strategy implementer, aims to optimize the performance indicators of the power grid, such as stability, efficiency, cost, etc., by dynamically adjusting the fractal control parameters.

[0225] Status t ∈S, represents the state of the power grid at time step t, including all necessary observation variables, such as V t , P t , Q t 、f t They represent voltage, active power, reactive power and frequency respectively.

[0226] Action a t ∈A, the agent is in state s t The control action taken under the control is to adjust the control parameters, such as adjusting the tap position of the substation transformer and the generator output.

[0227] Reward t The feedback obtained immediately after the agent performs an action reflects the immediate impact of the action on the performance of the power grid; the reward function design needs to consider long-term stability and efficiency, and may combine multiple performance indicators, such as reducing power loss and improving frequency stability.

[0228] The goal is to maximize the cumulative reward through the learning process, that is, to find the optimal strategy, which is expressed as:

[0229]

[0230] In the formula, π*(s) is the optimal strategy π* under a given state s; the optimal strategy refers to the strategy that can maximize the expected cumulative reward in any state; argmax π To find the strategy that maximizes the expression in brackets among all strategies π; is the expected value under strategy π, which is the average or expected sum of future rewards when strategy π is executed, taking into account all possible behavior sequences and their corresponding reward distributions during the strategy execution process; t is the time step, starting from 0 and ending in infinity; r t is the immediate reward obtained at time step t; γ is a discount factor, whose value is between 0 and 1, which is used to reduce the importance of future rewards; when γ is close to 1, the future reward is almost equivalent to the immediate reward; when γ is close to 0, more emphasis is placed on the immediate reward and the long-term impact is ignored.

[0231] The implementation process includes:

[0232] Initialize parameters, set the initial strategy π0, define the state space S, action space A, reward function R(s, a) and discount factor γ.

[0233] At each time step t, the agent observes the current state of the grid s t .

[0234] The agent follows the current policy π t , select action a t =π t (s t ), and apply it to the power grid, observing the next state s t+1 and instant rewards t .

[0235] Use Q-learning algorithm according to s t , a t , r t ,s t+1 Update strategy π t , optimize long-term rewards.

[0236] The learned strategies are directly converted into dynamic adjustments to the fractal control parameters. This process continues during the operation of the power grid. The agent continuously learns and adapts to new states and environmental changes, thereby achieving adaptive optimization of control parameters.

[0237] In this embodiment, the historical operation data of the power grid is first collected to provide data support for all subsequent analysis and control strategies. By obtaining the actual operation conditions of the power grid, such as frequency fluctuations, output of distributed energy resources, and external disturbance information, an empirical basis is provided for building models and predictions.

[0238] Based on the data collected by S1, a fractal chaos attractor model is constructed, and the fractal dimension and correlation coefficient of the power grid state variables are calculated. This is a high-level abstraction of the dynamic characteristics of the power grid, which identifies the self-similarity and complexity characteristics in the dynamic behavior of the power grid, and lays a theoretical foundation for understanding and predicting the behavior of the power grid.

[0239] The Lyapunov exponent analysis method is applied to quantify the stability of the system, predict the stability boundary of the power grid, and identify the key to the potential instability trend of the system, so that the control strategy can intervene before the problem occurs.

[0240] According to the instability trend identified by S3 and the fractal characteristics of the power grid in S2, the designed control signal has a self-similar structure that matches the natural fluctuation pattern of the power grid, which enhances the compatibility of the control measures with the power grid dynamics and improves the control efficiency.

[0241] The control signal designed by S4 is injected into the distributed energy control system in this step, and synchronization between different nodes is achieved through specific communication protocols and algorithms, which improves the overall responsiveness of the system and strengthens its resistance to disturbances and self-repair capabilities.

[0242] The control strategy implemented in S5 is optimized in S6. According to the real-time status of the power grid and changes in the external environment, the fractal control parameters are dynamically adjusted to ensure that the control strategy always matches the actual situation, thus achieving the adaptability of the control strategy.

[0243] Example 2

[0244] like Figure 2 As shown, the present invention discloses an adaptive smart grid control system, the system comprising:

[0245] The data collection module 10 is used to collect historical operation data of the power grid.

[0246] The model building module 20 is used to build a fractal chaos attractor model and calculate the fractal dimension of the power grid state variable.

[0247] The boundary prediction module 30 predicts the stability boundary of the power grid based on the fractal chaos attractor model.

[0248] The signal generation module 40 designs a fractal control signal and generates a control signal with a self-similar structure.

[0249] The system control module 50 is used to inject the fractal control signal into the distributed energy control system.

[0250] The parameter optimization module 60 dynamically adjusts the fractal control parameters according to the power grid operation status and external environment changes.

[0251] As an optional implementation, the model building module 20 of the present invention specifically includes:

[0252] Select key state variables and arrange the selected variables in chronological order to form a sequence.

[0253] Map the one-dimensional time series to a high-dimensional space and construct a state vector sequence, including:

[0254] Embedding parameters are selected, including an embedding dimension m and a time delay τ.

[0255] For each time point t, a state vector is constructed, expressed as:

[0256] X t =[f(t), f(t+τ),..., f(t+(m-1)τ), V(t), V(t+τ),..., V(t+(m-1)τ)]

[0257] For each time point t, based on the above m and τ, construct a state vector X containing the past m-1 time points and the current state t ; Through m and τ, the original one-dimensional time series is converted into a set of points in the multidimensional state space, that is, each time point corresponds to an m-dimensional vector.

[0258] Calculate the correlation dimension D C , specifically including:

[0259] For any two state vectors X i , X j , calculate the distance between them, expressed as:

[0260]

[0261] In the formula, x ik and x jk They are the state vector X i and X j The elements in the kth dimension.

[0262] Count the number of neighborhood point pairs, set the neighborhood radius r, and count the state vector X of each point i The number of point pairs N in the neighborhood of i (r); that is, X i The number of state vectors within a sphere with a center and a radius of r.

[0263] Find the average number of neighborhood point pairs of all points and get N(r), which is expressed as:

[0264]

[0265] Where N is the total number of state vectors; N(r) describes how the average distribution density of points in the state space changes with radius.

[0266] Plot the correlation function and calculate the correlation dimension, expressed as:

[0267] The correlation function C(r) is defined as the natural logarithm of the logarithm of the neighborhood points, that is, C(r) = ln[N(r)].

[0268] The slope D obtained by linear regression analysis c , expressed as:

[0269]

[0270] Where D c is the correlation dimension.

[0271] As an optional implementation, the boundary prediction module 30 of the present invention specifically includes:

[0272] The system is locally linearized and approximated as the Jacobian matrix J(X t , t), where X is the state vector and t is the time.

[0273] Perform iterative operations, including:

[0274] Initial vector setting, choose a small initial vector difference δX0.

[0275] Iterative update, at each time step t, by δX t+1 = J(X t , t)·δX t Update vector difference.

[0276] Average growth rate calculation, calculate the average growth rate Λ over a period of time I , expressed as:

[0277]

[0278] Where T is the length of time for analysis.

[0279] Repeating the iterative process for different orthogonal directions in the system state space, we obtain a series of Lyapunov exponents λ1, λ2, ..., λ n , forming a Lyapunov spectrum; if the maximum Lyapunov exponent λ max > 0, indicating that the system is in a chaotic state and has potential instability; if λ max <0, the system tends to be stable; by continuously tracking the maximum Lyapunov exponent λ max , find λ max= 0, identifying the critical point from stability to chaos, i.e., the stability boundary.

[0280] As an optional implementation, the signal generating module 40 of the present invention specifically includes:

[0281] Using the fBm fractal model, Hurst exponent and time scale, the control signal is generated and expressed as:

[0282]

[0283] In the formula, B H (t) is the fBm value at time point t; t K =KΔt, Δt is the time interval; σ K is the standard deviation of each step based on the Hurst exponent adjustment; Z K is a standard normal random distributed variable.

[0284] The generated control signal is fine-tuned by modifying its amplification factor or using a filter.

[0285] As an optional implementation mode, the parameter optimization module 60 of the present invention specifically includes:

[0286] Define reinforcement learning parameters, including:

[0287] Status t ∈S, represents the state of the power grid at time step t; action a t ∈A, the agent is in state s t The control action taken under the condition is to adjust the control parameters; the reward r t , the feedback the agent gets immediately after performing an action.

[0288] Maximizing the cumulative reward through the learning process, that is, finding the optimal strategy, is expressed as:

[0289]

[0290] Where π*(s) is the optimal strategy π* in a given state s; argmax π To find the strategy that maximizes the expression in brackets among all strategies π; is the expected value under strategy π, which is the average or expected sum of future rewards when strategy π is executed; t is the time step; r t is the immediate reward obtained at time step t; γ is the discount factor.

[0291] The implementation process includes:

[0292] Initialize parameters, set the initial strategy π0, define the state space S, action space A, reward function R(s, a) and discount factor γ.

[0293] At each time step t, the agent observes the current state of the grid s t .

[0294] The agent follows the current policy π t , select action a t =π t (s t ), and apply it to the power grid, observing the next state s t+1 and instant rewards t .

[0295] Use Q-learning algorithm according to s t , a t , r t ,s t+1 Update strategy π t , optimize long-term rewards.

[0296] The learned policy is directly converted into a dynamic adjustment of the fractal control parameters.

[0297] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An adaptive smart grid control method, characterized in that: The method comprises: Step S1: Collect historical power grid operation data; Step S2: construct a fractal chaos attractor model and calculate the fractal dimension of the power grid state variable; Step S3: predicting the stability boundary of the power grid based on the fractal chaotic attractor model; Step S4: designing a fractal control signal to generate a control signal with a self-similar structure; Step S5: injecting the fractal control signal into the distributed energy control system; Step S6: dynamically adjust the fractal control parameters according to the grid operation status and external environment changes.

2. The adaptive smart grid control method according to claim 1, characterized in that: The fractal chaotic attractor model is constructed to calculate the fractal dimension of the power grid state variable, specifically including: Select key state variables and arrange the selected variables in chronological order to form a sequence; Map the one-dimensional time series to a high-dimensional space and construct a state vector sequence, including: Selecting embedding parameters, the embedding parameters comprising an embedding dimension m and a time delay τ; For each time point t, a state vector is constructed, expressed as: X t =[f(t),f(t+τ),...,f(t+(m-1)τ),V(t),V(t+τ),...,V(t+(m-1)τ)] For each time point t, based on the above m and τ, construct a state vector X containing the past m-1 time points and the current state t ; Through m and τ, the original one-dimensional time series is converted into a point set in the multidimensional state space, that is, each time point corresponds to an m-dimensional vector; Calculate the correlation dimension D C , specifically including: For any two state vectors X i , X j , calculate the distance between them, expressed as: In the formula, x ik and x jk They are the state vector X i and X j Elements in the kth dimension; Count the number of neighborhood point pairs, set the neighborhood radius r, and count the state vector X of each point i The number of point pairs N in the neighborhood of i (r); that is, X i The number of state vectors in a sphere with a center and a radius of r; Find the average number of neighborhood point pairs of all points and get N(r), which is expressed as: Where N is the total number of state vectors; N(r) describes how the average distribution density of points in the state space changes with radius; Plot the correlation function and calculate the correlation dimension, expressed as: Define the correlation function C(r) as the natural logarithm of the logarithm of the neighborhood points, that is, C(r) = ln[N(r)]; The slope D obtained by linear regression analysis C , expressed as: Where D C is the correlation dimension.

3. The adaptive smart grid control method according to claim 1, characterized in that: The predicting of the stability boundary of the power grid based on the fractal chaotic attractor model specifically includes: The system is locally linearized and approximated as the Jacobian matrix J(X t , t), where X is the state vector and t is the time; Perform iterative operations, including: Initial vector setting, select a small initial vector difference δX0; Iterative update, at each time step t, by δX t+1 = J(X t , t)·δX t Update vector difference; Average growth rate calculation, calculate the average growth rate δ over a period of time I , expressed as: In the formula, T is the length of time for analysis; Repeating the iterative process for different orthogonal directions in the system state space, we obtain a series of Lyapunov exponents λ1, λ2, ..., λ n , forming a Lyapunov spectrum; if the maximum Lyapunov exponent λ max > 0, indicating that the system is in a chaotic state and has potential instability; if λ max <0, the system tends to be stable; by continuously tracking the maximum Lyapunov exponent λ max , find λ max = 0, identifying the critical point from stability to chaos, i.e., the stability boundary.

4. The adaptive smart grid control method according to claim 1, characterized in that: The design of the fractal control signal to generate a control signal with a self-similar structure specifically includes: Using the fBm fractal model, Hurst exponent and time scale, the control signal is generated and expressed as: In the formula, B H (t) is the fBm value at time point t; t K =KΔt, Δt is the time interval; σ K is the standard deviation of each step based on the Hurst exponent adjustment; Z K is a variable with standard normal random distribution; The generated control signal is fine-tuned by modifying its amplification factor or using a filter.

5. The adaptive smart grid control method according to claim 1, characterized in that: According to the operation status of the power grid and changes in the external environment, the fractal control parameters are dynamically adjusted, including: Define reinforcement learning parameters, including: Status t ∈S, represents the state of the power grid at time step t; Action a t ∈A, the agent is in state s t The control action taken under the above conditions is to adjust the control parameters; Reward t , the feedback the agent gets immediately after performing an action; Maximizing the cumulative reward through the learning process, that is, finding the optimal strategy, is expressed as: In the formula, π * (s) is the optimal strategy π in a given state s * ; argmax π To find the strategy that maximizes the expression in brackets among all strategies π; is the expected value under strategy π, which is the average or expected sum of future rewards when strategy π is executed; t is the time step; r t is the immediate reward obtained at time step t; γ is the discount factor; The implementation process includes: Initialize parameters, set the initial strategy π0, define the state space S, action space A, reward function R(s, a) and discount factor γ; At each time step t, the agent observes the current state of the grid s t ; The agent follows the current policy π t , select action a t =π t (s t ), and apply it to the power grid, observing the next state s t+1 and instant rewards t ; Use Q-learning algorithm according to s t , a t , r t ,s t+1 Update strategy π t , optimize long-term rewards; The learned policy is directly converted into a dynamic adjustment of the fractal control parameters.

6. An adaptive smart grid control system, characterized in that: The system comprises: Data collection module, used to collect historical operation data of power grid; A model building module is used to build a fractal chaos attractor model and calculate the fractal dimension of the power grid state variables; A boundary prediction module, which predicts the stability boundary of the power grid based on the fractal chaotic attractor model; Signal generation module, designs fractal control signals and generates control signals with self-similar structures; A system control module for injecting fractal control signals into a distributed energy control system; The parameter optimization module dynamically adjusts the fractal control parameters according to the grid operation status and external environment changes.

7. The adaptive smart grid control system according to claim 6, characterized in that: The model building module specifically includes: Select key state variables and arrange the selected variables in chronological order to form a sequence; Map the one-dimensional time series to a high-dimensional space and construct a state vector sequence, including: Selecting embedding parameters, the embedding parameters comprising an embedding dimension m and a time delay τ; For each time point t, a state vector is constructed, expressed as: X t =[f(t),f(t+τ),...,f(t+(m-1)τ),V(t),V(t+τ),...,V(t+(m-1)τ)] For each time point t, based on the above m and τ, construct a state vector X containing the past m-1 time points and the current state t ; Through m and τ, the original one-dimensional time series is converted into a point set in the multidimensional state space, that is, each time point corresponds to an m-dimensional vector; Calculate the correlation dimension D C , specifically including: For any two state vectors X i , X j , calculate the distance between them, expressed as: In the formula, x ik and x jk They are the state vector X i and X j Elements in the kth dimension; Count the number of neighborhood point pairs, set the neighborhood radius r, and count the state vector X of each point i The number of point pairs N in the neighborhood of i (r); that is, X i The number of state vectors in a sphere with a center and a radius of r; Find the average number of neighborhood point pairs of all points and get N(r), which is expressed as: Where N is the total number of state vectors; N(r) describes how the average distribution density of points in the state space changes with radius; Plot the correlation function and calculate the correlation dimension, expressed as: Define the correlation function C(r) as the natural logarithm of the logarithm of the neighborhood points, that is, C(r) = ln[N(r)]; The slope D obtained by linear regression analysis C , expressed as: Where D C is the correlation dimension.

8. The adaptive smart grid control system according to claim 6, characterized in that: The boundary prediction module specifically includes: The system is locally linearized and approximated as the Jacobian matrix J(X t , t), where X is the state vector and t is the time; Perform iterative operations, including: Initial vector setting, select a small initial vector difference δX0; Iterative update, at each time step t, by δX t+1 = J(X t , t)·δX t Update vector difference; Average growth rate calculation, calculate the average growth rate δ over a period of time I , expressed as: Where T is the length of time for analysis; Repeating the iterative process for different orthogonal directions in the system state space, we obtain a series of Lyapunov exponents λ1, λ2, ..., λ n , forming a Lyapunov spectrum; if the maximum Lyapunov exponent λ max > 0, indicating that the system is in a chaotic state and has potential instability; if λ max <0, the system tends to be stable; by continuously tracking the maximum Lyapunov exponent λ max , find λ max = 0, identifying the critical point from stability to chaos, i.e., the stability boundary.

9. The adaptive smart grid control system according to claim 6, characterized in that: The signal generating module specifically comprises: Using the fBm fractal model, Hurst exponent and time scale, the control signal is generated and expressed as: In the formula, B H (t) is the fBm value at time point t; t K =KΔt, Δt is the time interval; σ K is the standard deviation of each step based on the Hurst exponent adjustment; Z K is a variable with standard normal random distribution; The generated control signal is fine-tuned by modifying its amplification factor or using a filter.

10. The adaptive smart grid control system according to claim 6, characterized in that: The parameter optimization module specifically includes: Define reinforcement learning parameters, including: Status t ∈S, represents the state of the power grid at time step t; action a t ∈A, the agent is in state s t The control action taken under the condition is to adjust the control parameters; the reward r t , the feedback the agent gets immediately after performing an action; Maximizing the cumulative reward through the learning process, that is, finding the optimal strategy, is expressed as: Where π*(s) is the optimal strategy π* in a given state s; argmax π To find the strategy that maximizes the expression in brackets among all strategies π; is the expected value under strategy π, which is the average or expected sum of future rewards when strategy π is executed; t is the time step; r is the immediate reward obtained at time step t; γ is the discount factor; The implementation process includes: Initialize parameters, set the initial strategy π0, define the state space S, action space A, reward function R(s, a) and discount factor γ; At each time step t, the agent observes the current state of the grid s t ; The agent follows the current policy π t , select action a t =π t (s t ), and apply it to the power grid, observing the next state s t+1 and instant rewards t ; Use the Q-leaming algorithm to calculate the t , a t , r t ,s t+1 Update strategy π t , optimize long-term rewards; The learned policy is directly converted into a dynamic adjustment of the fractal control parameters.

Images