Frequency modulation demand quantification and retention evaluation method and system based on Kriging agent model

Abstract

The invention discloses a frequency modulation demand quantification and retention evaluation method and system based on a Kriging agent model, and relates to the field of uncertainty quantification of a power system, and the method comprises the steps: collecting date features, meteorological factors and actual operation data of a regional power grid, generating a large number of Monte Carlo simulation scenes, extracting and constructing fluctuation feature vectors, and carrying out the screening of the feature vectors, thereby achieving the quantification of a frequency modulation demand. The method comprises the steps of obtaining typical feature vectors input into Kriging agent model training, outputting a mean value and a standard deviation of net load prediction, and selecting proper distribution to obtain different confidence intervals so as to meet the confidence intervals of opportunity constraints to evaluate and quantify frequency modulation capacity retention. Compared with the prior art, the method has the advantages that the input-output data mapping relation is established based on the samples, the relation between active power fluctuation and frequency modulation resource quantification is effectively described, rapid fluctuation of new energy under the energy transformation background is considered, fine configuration of the frequency modulation capacity can be guided, and the safety and economical efficiency of the system are improved.

Description

Technical Field

[0001] This invention relates to the quantification of uncertainty in power systems, specifically to a reliable quantification method and system for frequency regulation capacity considering rapid fluctuations in new energy sources. Background Technology

[0002] The large-scale integration of renewable energy sources such as wind and solar power, while increasing the proportion of clean energy, also poses a severe challenge to the frequency stability of the power system due to their inherent intermittency, volatility, and uncertainty. Frequency is a key indicator for measuring power quality and the safe and stable operation of the system, and its stability is directly related to the real-time balance between power generation and consumption. To balance the active power deviation of the system caused by factors such as rapid fluctuations in renewable energy, it is necessary to reserve appropriate secondary frequency regulation capacity for AGC units in the day-ahead period to achieve precise scheduling of AGC units during real-time operation. Accurately assessing the frequency regulation capacity demand for each scheduling period is not only the cornerstone of maintaining frequency stability but also a fundamental prerequisite for optimizing resource allocation and improving the overall economic efficiency of the system. Existing frequency regulation capacity quantification methods suffer from problems such as unclear frequency regulation correlation characteristics, complex frequency regulation capacity characteristics leading to large sample requirements, and the difficulty in overcoming the risk of insufficient frequency regulation reservation. Currently, the industry still largely relies on experience to determine frequency regulation capacity demand. In summary, it is necessary to propose a data-driven reliable quantification method for frequency regulation capacity that considers the rapid fluctuations in renewable energy to improve the precision of power system frequency regulation capacity configuration under the new circumstances. Summary of the Invention

[0003] This invention addresses the problems of unclear frequency regulation correlation characteristics, complex frequency regulation capacity characteristics leading to large sample requirements, and the difficulty in overcoming the risk of insufficient frequency regulation reservation in existing frequency regulation capacity quantification methods. It proposes a data-driven, reliable frequency regulation capacity quantification method that considers the rapid fluctuations in renewable energy sources, thereby improving the precision of power system frequency regulation capacity allocation under new conditions. By collecting date characteristics, meteorological factors, and actual regional power grid operation data, the method inputs these data into a data-driven prediction model, outputting the mean and standard deviation of the net load forecast. Appropriate distributions are selected to obtain different confidence intervals, which satisfy chance-constrained confidence intervals for evaluating and quantifying frequency regulation capacity reservation, thus solving the problems mentioned in the background.

[0004] To achieve the above objectives, the present invention provides the following technical solution: This invention relates to a data-driven reliable quantification method for frequency regulation capacity that takes into account the rapid fluctuations of new energy sources. First, by collecting date characteristics, meteorological factors, and actual operation data of the regional power grid, a fluctuation feature vector is extracted and constructed. This vector is then filtered using the maximal information coefficient (MIC) method to obtain the typical feature vector input to the Kriging surrogate model for training. The mean and standard deviation of the net load forecast are output, and different confidence intervals are obtained by selecting a Gaussian distribution. The frequency regulation capacity retention is evaluated and quantified based on the confidence interval that satisfies the chance constraint.

[0005] The meteorological factors mentioned include: average temperature, average humidity, average light intensity, and average wind speed.

[0006] The aforementioned regional power grid actual operation data includes: historical load data, historical wind power generation data, historical photovoltaic power generation data, and historical net load power generation data.

[0007] The maximum information coefficient method refers to a method that uses mutual information theory to perceive deep, indirect, nonlinear correlations between variables and to measure the correlation between variables.

[0008] The Kriging surrogate model is a statistical, high-precision interpolation model that replaces the original complex simulation model, which is computationally expensive or time-consuming, with a mathematical model that has extremely low computational cost. Its mathematical essence is a Gaussian process regression model, and its core is to describe spatial correlation through the covariance function (or variation function) and on this basis, provide the optimal linear unbiased prediction and its uncertainty estimate for unknown points.

[0009] This invention also relates to a system for implementing the above method, comprising: a data acquisition unit, an input / output generation and prediction model construction unit, and a frequency regulation demand assessment and quantification unit; wherein: the data acquisition unit acquires date characteristics, meteorological factors, and actual operation data of the regional power grid; the input / output generation and prediction model construction unit extracts and constructs fluctuation feature vectors and filters them to obtain typical feature vectors for inputting the Kriging surrogate model for training, and outputs the mean and standard deviation of the net load forecast; a large number of Monte Carlo simulation scenarios are generated from historical data and daily forecast data, and appropriate distributions are selected to obtain different confidence intervals, so as to evaluate and quantify the frequency regulation capacity retention based on the confidence intervals that meet the chance constraints.

[0010] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention collects date characteristics, meteorological factors, and actual operation data of the regional power grid, inputs them into a data-driven prediction model, outputs the mean and standard deviation of the net load forecast, selects appropriate distributions to obtain different confidence intervals, and evaluates and quantifies frequency regulation capacity retention within confidence intervals that meet chance constraints. This invention establishes an input-output data mapping relationship based on samples, effectively describing the relationship between active power fluctuations and frequency regulation resource quantification, which is beneficial for guiding the refined allocation of frequency regulation capacity and improving system security and economy. Attached Figure Description

[0011] Figure 1 This is a flowchart of the present invention; Figure 2 A graph showing the up-modulation capacity and the maximum information coefficients of each explanatory variable; Figure 3 This is a multi-confidence level prediction interval chart for the maximum net load on a typical day; Figure 4 This represents the typical daily frequency regulation capacity and its percentage of the peak load for that day. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] like Figure 1 As shown, this embodiment relates to a frequency modulation demand quantification and retention assessment method based on the Kriging surrogate model, including: Step S1: Collect data on natural environmental factors and actual power system data, including date characteristics, average temperature, average humidity, average light intensity, average wind speed, historical load data, historical wind power generation data, historical photovoltaic power generation data, and historical net load power generation data; Step S2: Extract the fluctuation feature vector from the actual operating data of the power system and extract the typical feature vector using the maximum information coefficient method. The specific process includes: Step S21: Based on historical load data with a granularity of 1 minute, each 5-minute period is considered as a time interval (there are 288 time intervals per day). The average output level, maximum variation, and maximum variation between adjacent times within the scheduling period are used as feature variables: in: The average power level of the load during the period; This represents the actual load value within the time period. This represents the number of points within the scheduling period. This represents the maximum degree of load variation during that period. This represents the maximum load change between adjacent time points within the time period.

[0014] The average output level, maximum variation range, and maximum variation between adjacent times are extracted from the actual historical output data of new energy sources in different time periods as fluctuation characteristics.

[0015] in: The average power level of new energy sources during the period; The actual output of new energy sources during the period; This represents the number of points within the scheduling period. This represents the maximum degree of change in new energy sources during this period; This represents the maximum output change of new energy sources at adjacent moments within the time period.

[0016] Net load fluctuation curve obtained by superimposing the output and load of new energy power generation The maximum and minimum net load values ​​for different time periods are extracted as target variables for prediction, while the average output level, maximum variation range, and maximum variation between adjacent time periods are used as external explanatory variables for volatility.

[0017] in: The average net load power level during the period; The actual net load output during the time period; This represents the number of points within the scheduling period. This represents the maximum change in net load during that period. This represents the maximum change in net load output between adjacent time points within the time period.

[0018] In summary, the input and output of the net load fluctuation prediction model are as follows: in, , These represent the maximum and minimum net load values ​​during the scheduling period, respectively.

[0019] Step S22: To improve prediction accuracy, the main factors affecting net load need to be identified. This invention uses the maximal information coefficient (MIC) method to measure the similarity between the target variable and different external explanatory variables: in, , Let them be two random variables; , They are respectively , The number of grids divided in the direction, and , This represents the upper limit for the number of grid cells. Mutual information value; , They are respectively , Marginal distribution probability; for , The joint distribution probability.

[0020] Step S3: Input the typical feature vectors obtained in the previous step into the Kriging surrogate model for training, and output the mean and standard deviation of the net load prediction. The specific process is as follows: Let the training dataset be... have The 3D design space is obtained through the previous fluctuation vector extraction. a set of sample points ,in, Numerical analysis is performed on each sample point to obtain the corresponding response value. based on and To construct a Kriging surrogate model for this problem, the Kriging model needs to be built twice, separately, to predict the maximum and minimum net load values.

[0021] The Kriging model is an interpolation model whose interpolation result is defined as a linear weighted sum of known sample function response values, i.e.: To calculate the weighting coefficients By introducing statistical assumptions, the unknown function is considered as a concrete realization of a certain Gaussian static stochastic process. This static stochastic process is defined as follows: in, The unknown constant, also known as the global trend model, represents The expected value of the mathematical expression; With a mean of 0 and a variance of ( These random variables are static stochastic processes. At different locations in the design space, these random variables exhibit a certain correlation (or covariance). This covariance can be expressed as: in, It is a "correlation function" (related only to spatial distance), and satisfies the following conditions: equal to 1 when the distance is zero; equal to 0 when the distance is infinite; the correlation decreases as the distance increases.

[0022] Based on the above assumptions, the Kriging model seeks the optimal weighting coefficients. This makes the mean squared error Minimum, and satisfying the following interpolation conditions (or unbiased conditions): Using the Lagrange multiplier method, the optimal weighting coefficients The following system of linear equations gives in, ; Let be the Lagrange multiplier. Equation (52) can be written in matrix form as follows: in: in, The "correlation matrix" is composed of the correlation function values ​​between all known sample points; The "correlation vector" consists of the correlation function values ​​between the unknown point and all known sample points. Solving the linear equation system (52) and substituting it into equation (47) yields the Kriging model prediction. By inverting the block matrix, the model can be finally written as follows: Meanwhile, the mean squared error of the estimated value is estimated as follows: In the Kriging model shown in equation (57), the choice and calculation of the correlation function determine... and The construction of this will affect the final prediction accuracy, specifically as follows: in, , For any two different locations in the design space; These are hyperparameters. The relevant function of the Kriging surrogate model in this invention is selected as the Gaussian kernel function, whose expression is: Based on known training dataset Hyperparameters are obtained from the training data using the maximum likelihood estimation method. To obtain the optimal surrogate model, this invention uses an adaptive evolution algorithm based on the covariance matrix to maximize the likelihood function.

[0023] Step S4: Use the Kriging surrogate model to achieve a direct mapping between input and output, realize net load forecasting, and output the maximum net load fluctuation value for each time period. Minimum value The mean and variance. To ensure that the range formed by the predicted extreme values ​​of net load fluctuations effectively covers the actual net load fluctuation curve, frequency regulation demand needs to be quantified and assessed, including the following processes: Step S41: Based on historical actual system net load data, historical predicted net load data, and target day predicted net load system data, generate a large number of Monte Carlo simulation scenarios. First, calculate the historical prediction error. in, For the first sky Time period prediction error For the first sky Actual net load value for the period For the first sky Forecast net load for the period It is a set of historical days.

[0024] For each time period , in, For time period The mean error, For time period The standard deviation of the error.

[0025] Then for each scenario and each time period : in, Let be a standard normally distributed random variable. For the generated random error, For the target day's net load system forecast data, For the scene During the period The net load value.

[0026] Step S42: Assume that the extreme value uncertainty of net load fluctuation follows a Gaussian distribution, given a confidence level. Based on the generated random scenarios, count how many scenario curves satisfy the constraints of the net load prediction interval. That is, for all 288 scheduling periods, the chance constraint inequality must be satisfied: in, / This represents the predicted mean of the uncertainty of the maximum / minimum net load fluctuation during the scheduling period. For the scene During the period The net load value.

[0027] Adjust the confidence level and update the prediction interval until the constraints are met.

[0028] Step S5: Quantize the frequency modulation capacity requirement based on the prediction error confidence interval, as shown in the following formula: In the formula, for Initial net load value for the scheduling period; For confidence intervals Frequency modulation capacity requirements during scheduling periods; For confidence intervals Down-regulation capacity demand during scheduling periods; the purpose of the max / min function is to address the overall greater than 0 (i.e., the prediction error distribution of the maximum (minimum) value of net load fluctuations) ) or less than 0 (i.e. In extreme cases, the system does not need to reserve up- (down) frequency modulation capacity.

[0029] Finally, according to equation (65), the total frequency regulation capacity requirement for the scheduling period can be obtained, as shown in the following equation: in, for t Total frequency regulation capacity demand during the scheduling period.

[0030] Based on specific practical experiments, the above method was implemented in a real power grid in a certain region of China. The load and new energy data came from the actual data of the regional power grid from January to December 2024, a total of 366 days, with one point per minute, and 1440 points within one day. The fluctuation feature vectors of these data were extracted according to the method in step 2 to form the input data set with a scheduling period granularity of 5 minutes, and the output data set was the extreme value of net load fluctuation with a scheduling period granularity of 5 minutes. The constructed sample input and output feature vectors with a granularity of 5 minutes are shown in Table 1.

[0031] Table 1 Input and Output Vectors of the Kriging Proxy Model This implementation uses the MATLAB R2025a programming language, and the hardware environment is an Intel Core Ultra9 285H with 32.0GB of RAM.

[0032] To select the kernel function with the best predictive performance, a Gaussian process prediction model was trained using input-output samples obtained from 2016 sets of input data samples (data from the previous week). The prediction of the net load maxima on typical days was compared, and the mean absolute percentage error (MAPE) was used as the evaluation metric. The results are shown in Table 2. In the Kriging surrogate model, five kernel functions with nonlinear fitting capabilities were selected: Linear, Exponential, Matern-3_2, Matern-5_2, and Gaussian. Comparison revealed that the Gaussian kernel had a MAPE below 0.1%, achieving a better fitting effect. Therefore, the Gaussian kernel was chosen as the kernel function for the Kriging surrogate model in this example.

[0033] Table 2 Comparison of calculation errors for different kernel functions in the Kriging surrogate model This implementation example requires selecting feature data that is strongly correlated with the up-regulation capacity demand to form the input. The MIC coefficients between the maximum net load during a typical daily scheduling period and each explanatory variable are as follows: Figure 2 As shown in the figure, the average output of photovoltaic power generation, the maximum variation in photovoltaic power generation, the maximum adjacent variation in photovoltaic power generation, the average output of wind power generation, and the average output of the load are highly correlated with the demand for frequency regulation capacity; the maximum variation in load and the maximum adjacent variation in load are moderately correlated with the demand for frequency regulation capacity; and the maximum variation in wind power generation, the average level of net load, the maximum variation in net load, the maximum adjacent variation in net load, and the maximum adjacent variation in wind power generation are lowly correlated with the demand for frequency regulation capacity.

[0034] In this embodiment, the net load maximum value multi-confidence level interval estimation is as follows: Figure 3 As shown, over 288 scheduling periods, the average coverage of the 60% confidence interval was 87.15%, the average coverage of the 70% confidence interval was 90.28%, the average coverage of the 80% confidence interval was 93.40%, the average coverage of the 90% confidence interval was 96.18%, and the average coverage of the 98% confidence interval was 98.96%. The 98% confidence interval was ultimately selected.

[0035] In this embodiment, the calculated up-regulation capacity for 288 scheduling periods on a typical day and their respective percentages of the day's peak load are as follows: Figure 4 The maximum up-regulation capacity of the day was 36,351.55 MW, occurring at 7:30, accounting for 9.00% of the peak load of the day; the average percentage of up-regulation capacity to peak load was 1.01%.

[0036] Compared with existing technologies, this invention establishes an input-output data mapping relationship based on samples, effectively describing the relationship between active power fluctuations and frequency regulation resource quantification. This is beneficial for guiding the fine-grained configuration of frequency regulation capacity and improving the system's security and economy.

[0037] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0038] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for quantifying frequency modulation demand and evaluating retention based on the Kriging surrogate model, characterized in that, Includes the following steps: Step S1: Collect data on natural environmental factors and actual power system data; Step S2: Extract the fluctuation feature vector from the actual operating data of the power system and extract the typical feature vector using the maximum information coefficient method; Step S3: Input the obtained typical feature vectors into the Kriging surrogate model for training, and output the mean and standard deviation of the net load prediction; Step S4: Utilize the Kriging surrogate model to directly map the input and output, achieving net load forecasting. The output is the maximum net load fluctuation for each time period. Minimum value The mean and variance, Step S5: Quantize the frequency modulation capacity demand based on the prediction error confidence interval to obtain the total frequency modulation capacity demand for the scheduling period.

2. The method for frequency modulation demand quantification and retention evaluation based on the Kriging surrogate model according to claim 1, characterized in that, The collected data includes natural environmental factors and actual power system data, including date characteristics, average temperature, average humidity, average light intensity, average wind speed, historical load data, historical wind power generation data, historical photovoltaic power generation data, and historical net load power generation data.

3. The method for frequency modulation demand quantification and retention evaluation based on the Kriging surrogate model according to claim 2, characterized in that, Step S2 specifically includes: Step S21: Based on historical load data with a granularity of 1 minute, and dividing the data into 5-minute intervals, the average load output level, maximum variation, and maximum variation between adjacent times within each interval are used as feature variables. in: This represents the average load power level over the period. This represents the actual load value within the time period. The number of points within the scheduling period; This represents the maximum degree of load variation during that period. This represents the maximum load change between adjacent time points within the time period; The average output level, maximum variation range, and maximum variation between adjacent times are extracted from the actual historical output data of new energy sources for each time period as fluctuation characteristics: in: The average power level of new energy sources during the period; The actual output of new energy during the period; The number of points within the scheduling period; This represents the maximum degree of change in new energy sources during this period; This represents the maximum power output change of new energy sources at adjacent moments within the time period; Net load fluctuation curve obtained by superimposing the output and load of new energy power generation The maximum and minimum net load values ​​for different time periods are extracted as the target variables for prediction, while the average output level, maximum variation, and maximum variation between adjacent time periods are used as external explanatory variables for volatility. in: The average net load power level during the period; The actual net load output during the time period; The number of points within the scheduling period; This represents the maximum change in net load during that period. This represents the maximum change in net load output between adjacent time points within the time period; The input and output of the net load fluctuation prediction model are as follows: in, , These represent the maximum and minimum net load values ​​during the scheduling period, respectively. Step S22: Use the maximum information coefficient method to measure the similarity between the target variable and different external explanatory variables: in, , Let them be two random variables; , They are respectively , The number of grids divided in the direction, and , This represents the upper limit for the number of grid cells. Mutual information value; , They are respectively , Marginal distribution probability; for , The joint distribution probability.

4. The method for frequency modulation demand quantification and retention evaluation based on the Kriging surrogate model according to claim 3, characterized in that, Step S3 specifically includes: inputting the typical feature vectors obtained in step S22 into the Kriging surrogate model for training, and outputting the mean and standard deviation of the net load prediction. The specific process is as follows: Let the training dataset be... have The 3D design space is obtained through the previous fluctuation vector extraction. a set of sample points ,in, Numerical analysis is performed on each sample point to obtain the corresponding response value: based on and A Kriging surrogate model for this problem is constructed. For the prediction of the net load maxima and minima, the Kriging model is built separately in two steps. The Kriging model is an interpolation model whose interpolation result is defined as a linear weighted sum of known sample function response values, i.e.: To calculate the weighting coefficients By introducing statistical assumptions, the unknown function is considered as a concrete realization of a certain Gaussian static stochastic process, which is defined as follows: in, The unknown constant, also known as the global trend model, represents The expected value of the mathematical expression; With a mean of 0 and a variance of ( In a static stochastic process, these random variables exhibit covariance at different locations within the design space. The covariance is expressed as: in, This is the "correlation function", which is only related to spatial distance and satisfies the following conditions: it equals 1 when the distance is zero; it equals 0 when the distance is infinite; the correlation decreases as the distance increases. Kriging model seeks optimal weighting coefficients. This makes the mean squared error Minimum, and satisfying the following interpolation conditions: Using the Lagrange multiplier method, the optimal weighting coefficients The following system of linear equations gives in, ; For the Lagrange multiplier, equation (19) can be written in matrix form as follows: in, in, The "correlation matrix" is composed of the correlation function values ​​between all known sample points; The "correlation vector" is composed of the correlation function values ​​between the unknown point and all known sample points. Solving the linear equation system (19) and substituting it into equation (14) yields the Kriging model prediction. By inverting the block matrix, the model can be finally written as follows: Meanwhile, the mean squared error of the estimated value is estimated as follows: In the Kriging model shown in equation (23), the choice and calculation of the correlation function determine... and The construction of this will affect the final prediction accuracy, specifically as follows: in, , For any two different locations in the design space; For hyperparameters, the relevant function of the Kriging surrogate model is chosen to be the Gaussian kernel function, with the expression: Based on known training dataset Hyperparameters are obtained from the training data using the maximum likelihood estimation method. The optimal value is obtained by using the covariance matrix adaptive evolution algorithm to maximize the likelihood function and obtain the optimal surrogate model.

5. The method for frequency modulation demand quantification and retention evaluation based on the Kriging surrogate model according to claim 4, characterized in that, Step S4 specifically involves: Step S41: Based on the historical actual data of system net load, the historical predicted data of net load, and the system predicted data of net load on the target day, generate a large number of Monte Carlo simulation scenarios. First, calculate the historical prediction error: in, For the first sky Time period prediction error For the first sky Actual net load value for the period For the first sky Forecast net load for the period A set of historical days; For each time period , in, For time period The mean error, For time period The standard deviation of the error; Then for each scenario and each time period : in, Let be a standard normally distributed random variable. For the generated random error, For the target day's net load system forecast data, For the scene During the period Net load value; Step S42: Assume that the extreme value uncertainty of net load fluctuation follows a Gaussian distribution, given a confidence level. Based on the generated random scenarios, count how many scenario curves satisfy the constraints of the net load prediction interval, that is, for all 288 scheduling periods, the chance constraint inequality must be satisfied: in, / This is the predicted mean of the uncertainty of the maximum / minimum net load fluctuation during the scheduling period. For the scene During the period Net load value; Adjust the confidence level and update the prediction interval until the constraints are met.

6. The method for frequency modulation demand quantification and retention evaluation based on the Kriging surrogate model according to claim 5, characterized in that, In step S5, the frequency modulation capacity requirement is quantized based on the prediction error confidence interval, as shown in the following formula: In the formula, for Initial net load value for the scheduling period; For confidence intervals Frequency modulation capacity requirements during scheduling periods; For confidence intervals Down-regulation capacity requirements during the scheduling period; Finally, according to equation (32), the total frequency regulation capacity requirement for the scheduling period is obtained, as shown in the following equation. in, for Total frequency regulation capacity demand during the scheduling period.

7. A system for implementing the frequency modulation demand quantification and retention assessment method based on the Kriging surrogate model as described in any one of claims 1-6, characterized in that, include: The system comprises a data acquisition unit, an input / output generation and prediction model construction unit, and a frequency regulation demand assessment and quantification unit. Specifically, the data acquisition unit collects date characteristics, meteorological factors, and actual regional power grid operation data. The input / output generation and prediction model construction unit extracts and constructs fluctuation feature vectors and filters them to obtain typical feature vectors for training the Kriging surrogate model, outputting the mean and standard deviation of the net load forecast. The frequency regulation demand assessment and quantification unit generates a large number of Monte Carlo simulation scenarios from historical data and daily forecast data. It selects appropriate distributions to obtain different confidence intervals, and evaluates and quantifies the frequency regulation capacity retention based on the confidence intervals that meet the chance constraints.

Images