Power system regional secondary frequency modulation demand time-phased prediction method and related device

Abstract

The invention provides a power system regional secondary frequency modulation demand time-phased prediction method and a related device. The method comprises the following steps: acquiring regional load data, and predicting by using a super-resolution deep learning model to obtain regional high-resolution load data; the regional high-resolution load data is corrected by using the uncertainty of new energy power generation prediction; based on the corrected regional high-resolution load data, a rolling average method is adopted to calculate regional frequency modulation demand prediction quantity at each sampling moment; correcting the regional frequency modulation demand predictive quantity by using a control criterion of power system scheduling; and determining a demand prediction value of power system scheduling in different time periods based on the principle of statistics. According to the method, factors such as power grid operation data and new energy uncertainty are comprehensively analyzed, the secondary frequency modulation demand of the system can be dynamically distributed in different time periods, various resources are fully utilized during regional power system scheduling, and therefore the problems that an existing frequency modulation mechanism unit is too large in reserved allowance and low in economical efficiency and system operation efficiency are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and in particular relates to a time-division prediction method for regional secondary frequency regulation demand of a power system and related devices. Background Art

[0002] As a core component of achieving the dual carbon goals, the high integration of renewable energy is accelerating the decarbonization of the power system. However, the volatility and uncertainty of large-scale renewable energy integration also pose significant challenges to the stable operation of the power grid, particularly in frequency regulation. Traditional frequency regulation, which relies on large thermal power plants, is unable to meet the demands of the new power system. Therefore, forecasting regional secondary frequency regulation demand is particularly important in this context.

[0003] Accurate forecasting of secondary frequency regulation demand not only ensures grid frequency stability but also improves the economic efficiency and efficiency of power system dispatch. Combined with day-ahead market dispatch management, by predicting regional secondary frequency regulation demand by time period within the day, grid resource allocation can be optimized, unnecessary frequency regulation costs can be reduced, and the grid's response capacity can be enhanced. Especially with the increasing adoption of distributed energy, energy storage technology, and demand response, the potential for renewable energy to participate in grid frequency regulation is being further unleashed. However, the uncertainty of renewable energy output and the risk of curtailment also place higher demands on the accurate forecasting of regional load changes and frequency regulation demand.

[0004] The current regional secondary frequency regulation mechanism has several significant shortcomings. Most existing methods for forecasting frequency regulation demand fail to fully account for the high uncertainty of renewable energy generation and the complexity of load fluctuations. Some forecasting models simply extrapolate trends based on historical load data, ignoring the impact of renewable energy output fluctuations on load forecast accuracy. This results in significant deviations between forecast results and actual frequency regulation demand. Summary of the Invention

[0005] In view of this, the present invention provides a time-sharing prediction method and related devices for the regional secondary frequency regulation demand of the power system, aiming to accurately predict the time-sharing frequency regulation demand of the next day by comprehensively analyzing external factors such as power grid operation data and new energy uncertainty, thereby improving the operation safety and economy of the power grid, and providing important technical support for the power grid to achieve green and low-carbon goals.

[0006] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for predicting regional secondary frequency regulation demand in a power system by time period, comprising the following steps:

[0008] Collect regional load data and predict regional high-resolution load data using a pre-trained super-resolution deep learning model;

[0009] The uncertainty of renewable energy generation forecast is used to determine the fluctuation degree of load data forecast error, and the regional high-resolution load data is corrected based on the fluctuation degree of forecast error.

[0010] Based on the corrected regional high-resolution load data, the rolling average method is used to calculate the regional frequency regulation demand forecast at each sampling moment;

[0011] Use the control criteria of power system dispatch to revise the regional frequency regulation demand forecast;

[0012] The standard deviation of the regional frequency regulation demand forecast at all sampling moments in each dispatching period is calculated by time period, and the optimal standard deviation value is selected as the demand forecast value for power system dispatching in each dispatching period based on statistical principles. The optimal standard deviation value is the standard deviation value that covers the set proportional dispatching demand in the dispatching period.

[0013] Furthermore, the super-resolution deep learning model adopts a deep learning network model based on an encoder-decoder structure. The deep learning network model uses the Adam optimizer for model training. The training process is carried out by minimizing the predefined total loss function. The total loss function is as follows:

[0014]

[0015]

[0016]

[0017] Where, is the total loss function; is the reconstruction error function, For real high-resolution data, High-resolution data output for the model; is the regularization term, is the regularization coefficient.

[0018] Furthermore, the uncertainty of renewable energy generation forecast is used to determine the degree of fluctuation of load data forecast error, and the regional high-resolution load data is corrected based on the degree of fluctuation of the forecast error, including:

[0019] According to the uncertainty of forecasting various items in the new energy power generation forecast, the error distribution of various forecast items is determined;

[0020] For the error distribution of various forecast items, the Gumbel Copula function is used to establish a joint distribution;

[0021] Generate sample points of various prediction items from the joint distribution;

[0022] Calculate the net load error of the sample points and divide it into positive error samples and negative error samples; the calculation expression of the net load error is as follows:

[0023]

[0024] Where, is the net load error. When it is greater than zero, the corresponding net load error sample is divided into a positive error sample. When it is less than zero, the corresponding net load error sample is divided into a negative error sample. The sample point forecast value of the load in the forecast project; The sum of the predicted values ​​of the sample points of the power generation project in the forecast project;

[0025] The upper and lower standard deviations of the net load error are calculated based on the positive error samples and the negative error samples. The calculation expressions of the upper and lower standard deviations are as follows:

[0026]

[0027]

[0028] Where, and They are the upper standard deviation and the lower standard deviation, which are used to indicate the degree of fluctuation of the load data prediction error; and are the number of positive error samples and negative error samples, respectively. and are the i-th net load error sample in the positive error sample and the j-th net load error sample in the negative error sample respectively;

[0029] The regional high-resolution load data were corrected according to the upper and lower standard deviations as follows;

[0030]

[0031] Where, is the load data after correction at time t, is the load data at time t in the regional high-resolution load data, is the upper bound of the load forecast error, is the lower bound of the load forecast error.

[0032] Furthermore, sample points are generated from the joint distribution, including:

[0033] Sample the variables for each predictor item from a uniform distribution;

[0034] The uniformly distributed samples are transformed by the inverse Gumbel Copula function to generate samples with dependency;

[0035] Through the inverse cumulative distribution function of the marginal distribution, the generated Copula samples are mapped back to the original variable space to obtain the actual error samples of each predicted item, and the error samples are sample points.

[0036] Furthermore, the rolling average method is used to calculate the regional frequency modulation demand forecast at each sampling moment, including:

[0037] The load data of each load point in the corrected regional high-resolution load data is recorded as the first load data;

[0038] For each load point, take several load values ​​before and after and perform sliding average to obtain the load data of each load point after being processed by the rolling average method, and record it as the second load data;

[0039] The difference between the first load data and the second load data of each load point is used as an estimate of the secondary frequency regulation capacity required by the control area at the corresponding moment;

[0040] The difference between the first load data of two adjacent load points is used as the required load of the control area at the corresponding time. .

[0041] Furthermore, the control criteria of power system dispatch are used to correct the regional frequency regulation demand forecast, including:

[0042] According to the BAAL criterion, determine the regional control deviation at time t The upper and lower limit conditions are met; the upper and lower limit values ​​are determined according to the following formula:

[0043]

[0044]

[0045] Where, and Regional control deviation The upper and lower limits of B are the frequency deviation coefficient of the control area. and are the upper and lower reference frequencies, To correct the frequency deviation, is the frequency deviation at time t; the calculation formula for correcting the frequency deviation is as follows:

[0046]

[0047] Where, is the marginal change in power, is the frequency response coefficient of the system;

[0048] Control deviations by region and frequency deviation The secondary frequency regulation capacity estimate is revised as follows:

[0049]

[0050] Where, It is the estimated value of secondary frequency regulation capacity demand in the control area determined by considering the BAAL criterion.

[0051] Furthermore, based on statistical principles, the optimal standard deviation value is selected as the demand forecast value for power system dispatch in each dispatch period, including:

[0052] Pick is the optimal standard deviation value; is the standard deviation of the regional frequency regulation demand forecast;

[0053] The optimal standard deviation value of each dispatching period within the day is calculated in turn as the demand forecast value of power system dispatching, and the time-based forecast result of the secondary frequency regulation demand within the day is obtained.

[0054] In a second aspect, the present invention provides a time-division forecasting device for regional secondary frequency regulation demand in a power system, comprising:

[0055] The data acquisition and processing module is used to collect regional load data and predict regional high-resolution load data through a pre-trained super-resolution deep learning model;

[0056] A first correction module is used to determine the degree of fluctuation of the load data prediction error by utilizing the uncertainty of the new energy power generation prediction, and to correct the regional high-resolution load data based on the degree of fluctuation of the prediction error;

[0057] The first prediction module is used to calculate the regional frequency regulation demand forecast at each sampling moment using a rolling average method based on the corrected regional high-resolution load data;

[0058] A second correction module is used to correct the regional frequency regulation demand forecast using the control criterion of power system dispatch;

[0059] The second prediction module is used to calculate the standard deviation of the regional frequency regulation demand forecast at all sampling moments in each scheduling period by time period, and select the best standard deviation value as the demand forecast value of the power system scheduling in each scheduling period based on statistical principles. The best standard deviation value is the standard deviation value covering the set proportional scheduling demand in the scheduling period.

[0060] In a third aspect, the present invention provides a computer device, comprising a processor and a memory:

[0061] The memory is used to store computer programs and send instructions of the computer programs to the processor;

[0062] The processor executes a time-division prediction method for regional secondary frequency regulation demand of a power system according to the instructions of the computer program as described in the first aspect.

[0063] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, a method for predicting the regional secondary frequency regulation demand of a power system according to the first aspect is implemented.

[0064] In summary, the present invention provides a method and related device for predicting the regional secondary frequency regulation demand of a power system by time period, including collecting regional load data and predicting regional high-resolution load data through a pre-trained super-resolution deep learning model; using the uncertainty of new energy power generation prediction to determine the degree of fluctuation of the load data prediction error, and correcting the regional high-resolution load data based on the degree of fluctuation of the prediction error; based on the corrected regional high-resolution load data, using the rolling average method to calculate the regional frequency regulation demand forecast at each sampling moment; using the control criterion of power system dispatch to correct the regional frequency regulation demand forecast; calculating the standard deviation of the regional frequency regulation demand forecast at all sampling moments in each dispatch period by time period, and selecting the optimal standard deviation value as the demand forecast value for power system dispatch in each dispatch period based on statistical principles, and the optimal standard deviation value covering the standard deviation value of the set proportion dispatch demand in the dispatch period. The present invention accurately predicts the time-of-use frequency regulation demand of the next day by comprehensively analyzing external factors such as power grid operation data and new energy uncertainty, thereby improving the operational safety and economy of the power grid, and providing important technical support for the power grid to achieve green and low-carbon goals. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 A flowchart of a method for predicting regional secondary frequency regulation demand in a power system by time period provided by an embodiment of the present invention;

[0067] Figure 2 A schematic diagram of a super-resolution deep learning network model provided by an embodiment of the present invention;

[0068] Figure 3A flow chart of frequency modulation demand prediction provided by an embodiment of the present invention;

[0069] Figure 4 A block diagram of a device for predicting regional secondary frequency regulation demand in a power system by time period provided by an embodiment of the present invention;

[0070] Figure 5 A block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purposes, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] See also Figure 1 This embodiment provides a method for predicting regional secondary frequency regulation demand in a power system by time period, including the following steps:

[0073] S1: Collect regional load data and predict regional high-resolution load data using a pre-trained super-resolution deep learning model.

[0074] It should be noted that regional load data refers to the data records of the power consumption of various electrical equipment in the power system of a specific area. It is usually collected at a certain time interval. For example, the total power consumption value of the area is recorded every 15 minutes, reflecting the changes in the power load level of the area over time.

[0075] The super-resolution deep learning model is a model built based on deep learning algorithms. It aims to convert low-resolution data (usually data with large time intervals and relatively insufficient data details, such as low-frequency sampled load data) into high-resolution data (data with smaller time intervals and more detailed changes, similar to the more detailed load change curve reflected by high-frequency sampled data).

[0076] S2: Use the uncertainty of renewable energy power generation forecasts to determine the degree of fluctuation of load data forecast errors, and correct the regional high-resolution load data based on the degree of fluctuation of the forecast errors.

[0077] It should be noted that new energy sources (such as wind power and photovoltaic power generation) are greatly affected by natural conditions (wind speed, light intensity, etc.), and their power generation is intermittent and fluctuating, resulting in a large error range when predicting their power generation.

[0078] This step takes into account the uncertainty inherent in renewable energy generation forecasts. Changes in the proportion of renewable energy in the power system can affect the overall load and the accuracy of forecasts. By analyzing the uncertainty in renewable energy generation forecasts, we quantify its impact on load data forecast errors. This means determining the degree of fluctuation in load data forecast errors. Based on this fluctuation, we then make appropriate corrections to the previously generated high-resolution regional load data. For example, if the error fluctuates significantly, we can adjust the data accordingly.

[0079] This step further optimizes and corrects the high-resolution load data by incorporating the influencing factor of renewable energy generation, reducing the forecast deviation caused by the uncertainty of renewable energy generation, making the load data more in line with the actual situation, and improving the reliability of subsequent frequency regulation demand forecasts.

[0080] S3: Based on the corrected regional high-resolution load data, the rolling average method is used to calculate the regional frequency regulation demand forecast at each sampling moment.

[0081] It should be noted that the rolling average method is a data processing method that calculates the average value of a certain number of continuous data in sequence. As the data is continuously updated, the calculated average interval also rolls forward accordingly, thereby smoothing data fluctuations and highlighting data change trends.

[0082] In this step, the corrected high-resolution regional load data is arranged in chronological order. An appropriate sampling window size is selected, and the regional frequency regulation demand forecast corresponding to each sampling moment is calculated using a rolling average method. For example, the data for every five consecutive moments is averaged, and then the average is continuously updated by moving backward one moment to obtain the forecast for each moment.

[0083] This step processes the data through the rolling average method, which can eliminate the random fluctuation interference in the load data to a certain extent, making the calculated regional frequency regulation demand forecast more stable and better reflecting the real demand change trend, which is convenient for subsequent further analysis and scheduling applications.

[0084] S4: Use the control criteria of power system dispatch to correct the regional frequency regulation demand forecast.

[0085] It should be noted that the control criteria for power system dispatch are a series of rules and standards followed in the operation of the power system, which are used to ensure the safe, stable and efficient operation of the power system. For example, the power balance criteria and voltage stability criteria will be adjusted according to the actual system status.

[0086] This step substitutes the regional frequency regulation demand forecast calculated previously into the control criteria framework for power system dispatching. Based on the conditions, restrictions, and adjustment mechanisms stipulated in these criteria, the forecast is modified accordingly to ensure that it fully complies with the requirements of the actual dispatching and operation of the power system.

[0087] This step ensures that the regional frequency regulation demand forecast is based on meeting the dispatching requirements such as power system safety and stability, avoids the situation where the prediction does not meet the actual operating specifications due to simple prediction, and enhances the practicality and guidance of the prediction results in actual dispatching work.

[0088] S5: Calculate the standard deviation of the regional frequency regulation demand forecast at all sampling moments in each dispatching period by time period, and select the best standard deviation value as the demand forecast value for power system dispatching in each dispatching period based on statistical principles. The best standard deviation value covers the standard deviation value of the set proportional dispatching demand in the dispatching period.

[0089] It should be noted that this step treats the regional frequency regulation demand forecasts at all sampling times within each dispatch period as a set of data, and calculates their standard deviations. Then, based on statistical principles, the optimal standard deviation is selected. This optimal standard deviation is defined as the standard deviation that covers the set proportion of dispatch demand within the dispatch period, that is, the standard deviation that reflects the data fluctuation characteristics at a reasonable level of coverage. This standard deviation is then used as the demand forecast for the power system dispatch during that dispatch period.

[0090] The method proposed in this embodiment first improves the resolution of load data using a super-resolution deep learning model. It then corrects the data by taking into account the uncertainty of renewable energy generation. It then smoothes the data using a rolling average method to obtain a preliminary frequency regulation demand forecast. This is then corrected to align with actual operation based on the power system's dispatch control criteria. Finally, a reasonable demand forecast for each dispatch period is determined using statistical standard deviations by time period. This completes a comprehensive chain from data acquisition and processing to final prediction. Each step works together to gradually optimize the prediction results, aiming to improve the accuracy and practicality of regional secondary frequency regulation demand forecasts in the power system and better guide power system dispatch. This method not only utilizes a deep learning model to improve data resolution but also fully considers the impact of renewable energy generation uncertainty, a key factor, on load data and subsequent frequency regulation demand forecasts, making targeted corrections. Compared to traditional prediction methods based solely on historical load factors, this method is more responsive to the complex realities of power systems. Furthermore, through multiple steps of data processing and forecast correction, from data quality improvement and preliminary prediction to optimization of dispatch criteria and final determination of forecast values ​​based on statistical principles, the accuracy and reliability of the forecast results are ensured across multiple dimensions. This distinguishes it from simple direct prediction models and makes it more adaptable to the complex and changing operating environments of power systems.

[0091] In one embodiment, the super-resolution deep learning model uses a deep learning network model based on an encoder-decoder structure. The principle of this model in predicting regional high-resolution load data is as follows:

[0092] Assume that the load time series of a certain area is , whose sampling frequency is Assumptions Is the length of Time series data, Indicates time The load value at that time point on the time series data contains the time period Assume that another data sequence Has the same physical quantity, but the sampling frequency is higher within the same time period, the frequency is ,in Is a positive integer, representing the super-resolution factor (SRF). The length is , and has more intensive time points. For example, the original data sequence samples a load point every 15 minutes, and the high-frequency data samples a load point every 15 seconds.

[0093] High-frequency sampling load data of power system It is possible to predict a more detailed load curve. However, it is difficult to obtain high-frequency data, and it is unrealistic for the power system dispatching agency to obtain high-frequency sampled load data in real time. Therefore, it is hoped that the existing low-frequency data can be used to (i.e., load data sampled every 15 minutes daily by the dispatching agency) to infer high-frequency data . and They are associated through the downsampling model, expressed as:

[0094]

[0095] in, is a downsampling matrix, The goal of the super-resolution problem (SRP) of load data is to find a mapping , so that the reconstructed high-frequency data It can restore the information lost during the downsampling process as much as possible.

[0096] Since the SRP problem is underdetermined, it means that there are countless possible high-frequency data sequences that satisfy the downsampling model. In order to solve this problem, it is placed in the maximum a posteriori probability (MAP) estimation framework to find the posterior probability The largest solution. According to Bayes' formula, the posterior probability can be expressed as:

[0097]

[0098] in is the likelihood function based on the downsampling model, yes The prior probability of is a constant. When given, it can be estimated by maximizing the following expression :

[0099]

[0100] This is equivalent to solving the following formula:

[0101]

[0102] in, This can be solved by modeling the downsampling process, and It is obtained by modeling prior information. The prior model acts as a regularization term to constrain the solution space so that the solution not only conforms to the observed data but also conforms to the prior distribution. Therefore, the MAP estimation problem can be expressed as:

[0103]

[0104] in, is a distortion measure based on the Gaussian noise assumption, is a regularization term that contains prior information about high-frequency data. This equation shows that is input and the downsampling matrix The MAP solution can be further expressed as:

[0105]

[0106] exist When fixed, it is equivalent to the constructed SRP mapping. This equation shows that the prior information is actually contained in the network parameter set middle.

[0107] The super-resolution deep learning network model constructed in this embodiment is used to realize the To high frequency load data The mapping adopts a deep learning network model based on the encoder-decoder structure. This model takes advantage of the convolutional neural network (CNN), residual blocks (Residual Blocks) and autoencoder (Autoencoder) to gradually extract features and reconstruct high-frequency load data. Its implementation structure includes four parts: upsampling layer, encoder, decoder, and information supplement layer. Input upsampling is responsible for upgrading low-frequency load data to high-frequency data; the feature extraction encoder is responsible for passing the upsampled data through the encoder's multi-layer convolution and pooling, gradually extracting features and compressing the time step; the feature reconstruction decoder is responsible for gradually restoring the temporal resolution of the high-dimensional feature map output by the encoder through the decoder's deconvolution layer, and supplementing the spatial information through jump connections, and finally generating a feature map with the same time step and resolution as the original data; the information supplement layer is responsible for passing the output of the decoder into the information supplement layer. After processing by the residual block, the expression ability of the feature map is further enhanced to generate the final high-resolution load forecast data. Its process image is as follows Figure 2 As shown in the figure, the specific implementation content of each layer is as follows:

[0108] a) Upsampling layer

[0109] The input 15-minute interval low-frequency data is increased to a 15-second high-frequency data through the upsampling layer by linear interpolation:

[0110]

[0111] The upsampled data has higher temporal resolution and is ready for subsequent feature extraction.

[0112] b) Encoder

[0113] The encoder extracts features through a series of convolutional layers and pooling layers, and gradually reduces the time step. Its structure is as follows:

[0114] First convolution layer: , converting the input time series data into a feature map with 16 channels.

[0115] SE Block: Reweights the features between channels of the feature maps of 16 channels to enhance important features.

[0116] Pooling layer: , reduce the time step and reduce the data dimension.

[0117] Second convolution layer: , expand the feature map to 32 channels.

[0118] SE Block: Reweights the features between channels on the feature maps of 32 channels.

[0119] Pooling layer: Use again Reduce the time step.

[0120] The third convolution layer: , expand the feature map to 64 channels.

[0121] SE Block: Reweights the features between channels on the feature maps of 64 channels.

[0122] Pooling layer: , continue to reduce the time step and compress the resolution of the time series.

[0123] The fourth convolution layer (bottleneck layer): , expand the feature map to 128 channels.

[0124] SE Block: Performs inter-channel feature reweighting on the feature map of 128 channels.

[0125] Pooling layer: , the final feature map has 128 channels.

[0126] Fifth convolution layer: , converting the feature map into a high-dimensional feature map with 256 channels.

[0127] c) Decoder

[0128] The decoder's function is to restore the high-dimensional feature maps extracted by the encoder to the time steps and number of channels (i.e., 1 channel) of the original input data. The decoder gradually restores the temporal and spatial resolution through a series of deconvolution layers and skip connections. Its structure is as follows:

[0129] First layer of deconvolution: , reducing the feature map from 256 channels to 128 channels, while increasing the time step through deconvolution operations with stride=2.

[0130] Skip connection: The fourth layer output of the corresponding encoder is added to the output of the deconvolution layer to help recover the lost spatial information.

[0131] Second layer of deconvolution: , reducing the feature map from 128 channels to 64 channels and continuing to increase the time step.

[0132] Skip connection: Add the third layer output of the corresponding encoder to the output of the deconvolution layer.

[0133] The third layer of deconvolution: , reducing the feature map from 64 channels to 32 channels and continuing to increase the time step.

[0134] Skip connection: Add the second layer output of the corresponding encoder to the output of the deconvolution layer.

[0135] The fourth layer of deconvolution: , the feature map is reduced from 32 channels to 16 channels and restored to the original time step.

[0136] Skip connection: Add the first layer output of the corresponding encoder to the output of the deconvolution layer.

[0137] Final convolutional layer: , the multi-channel feature map is converted into the final single-channel output through the convolution operation to generate the final high-resolution time series.

[0138] d) Information Supplementary Layer

[0139] Following the decoder, an information supplementation layer is included. Its primary task is to further process the high-resolution feature maps output by the decoder and enhance their expressiveness. This layer consists of a series of residual blocks, each containing two fully connected layers for global modeling and feature reconstruction. The residual block's input dimension is 480, corresponding to high-resolution feature maps for 240 time steps.

[0140] Residual block: Each residual block consists of two fully connected layers:

[0141] The first fully connected layer maps input features to output features using the ReLU activation function.

[0142] The second fully connected layer further processes the features and maintains the same feature dimension as the input.

[0143] Skip connection: The input features are directly added to the output features to form a residual structure, thereby alleviating the gradient vanishing problem.

[0144] Information supplementation: After processing by several residual blocks, the enhanced feature representation is obtained and used as the final output.

[0145] During model optimization, the Adam optimizer is used to minimize the defined loss function. The loss function during training includes the mean squared error (MSE) and a regularization term to constrain the model's solution space. The details are as follows:

[0146] a) Loss Function

[0147] The loss function consists of two parts:

[0148] Reconstruction error: Use mean square error (MSE) to measure the model output With real high-resolution data The difference between , defined as:

[0149]

[0150] Regularization term: In order to avoid overfitting, regularization term is added , which represents the high-resolution data Prior information constraints:

[0151]

[0152] in, is the regularization coefficient, which controls the weight of the regularization term.

[0153] b) Optimizer

[0154] The Adam optimizer is used for model training. The advantage of the Adam optimizer is its adaptive learning rate adjustment capability, which can achieve better convergence with less hyperparameter adjustment. Its parameter settings are as follows:

[0155] Learning rate ( ): 1 × 10e−4;

[0156] The first-order momentum term ( ): 0.9;

[0157] The second-order momentum term ( ): 0.999;

[0158] Attenuation term ( ): 1 × 10e−8;

[0159] The optimization process is performed by minimizing the following total loss function To carry out:

[0160]

[0161] In this section, the proposed model is trained and tested on a real-world load dataset, and its prediction performance is analyzed. The experimental results include common evaluation metrics such as MSE, MAPE, MAE, and RMSE. The model's effectiveness is also verified by visually comparing the actual and predicted load curves.

[0162] In one embodiment, the uncertainty of renewable energy generation forecast is used to determine the degree of fluctuation of load data forecast error, and regional high-resolution load data is corrected based on the degree of fluctuation of the forecast error, including:

[0163] S21: Based on the uncertainty of the forecast of various projects in the new energy power generation forecast, determine the error distribution of various forecast projects. Taking the forecast projects as load, photovoltaic and wind power as an example, calculate the distribution of wind power, photovoltaic and load forecast errors. Figure 3 This is a flowchart for frequency regulation demand forecasting based on data correction for photovoltaic and wind power. In forecasting load, wind power, and photovoltaic power generation, due to the influence of multiple uncertainties, the statistical characteristics of the errors often follow a specific probability distribution. The following are common assumptions and their mathematical representations:

[0164] a) Wind power forecast error distribution:

[0165] Wind power forecast errors are often assumed to follow a Weibull distribution. This is because wind speed is often modeled as a Weibull distribution, which in turn causes the wind power forecast error to follow this distribution. The probability density function of the Weibull distribution is defined as:

[0166]

[0167]

[0168] in, is the shape parameter, is the scale parameter, Represents the wind power prediction error.

[0169] b) Distribution of photovoltaic power generation prediction errors:

[0170] The photovoltaic power generation prediction error is usually assumed to obey distributed. Distribution is a probability distribution used to describe random variables in a finite interval and is suitable for modeling photovoltaic output forecast errors. The probability density function of the distribution is defined as:

[0171]

[0172] in, for function, and are the shape parameters, Represents the photovoltaic power generation power prediction error.

[0173] c) Load forecast error distribution:

[0174] Load forecast errors are usually assumed to follow a normal distribution (Gaussian distribution). This is because large-scale load data usually exhibits a central tendency, and the distribution of forecast errors can be reasonably assumed to be a symmetrical normal distribution. The probability density function of the normal distribution is defined as:

[0175]

[0176] in, is the mean, is the standard deviation, Represents the load forecast error.

[0177] S22: For the error distribution of various forecast items, the Gumbel Copula function is used to establish a joint distribution.

[0178] To estimate the correlation between different variables, this step uses the Gumbel Copula. A copula is a joint distribution function defined in the interval [0,1] and is suitable for describing the dependency structure of multiple variables. Its basic definition is:

[0179]

[0180] in, 、 and are the cumulative distribution functions of load, wind power and photovoltaic power generation, respectively, and is the copula parameter, describing the degree of dependence between variables. Gumbel Copula is often used to model extreme events, such as forecast errors in load, wind power, or photovoltaic power generation.

[0181] To estimate the parameter ϵ in the copula, the maximum likelihood estimation (MLE) method is used. By substituting the sampled values ​​into the copula function and solving the MLE, the optimal ϵ value is obtained.

[0182] Through Gumbel Copula, the variables can be calculated 、 、 The joint probability density function between . Joint probability density function is defined as follows:

[0183]

[0184] According to the chain rule, this formula can be further expanded as:

[0185]

[0186] The above formula can be further decomposed into the product of the derivatives of each marginal distribution

[0187]

[0188] in, The probability density function of the marginal distribution is expressed as follows. Using this formula, the probability density of the joint distribution can be obtained.

[0189] S23: Generate sample points of various prediction items from the joint distribution.

[0190] By the joint probability density function Sampling can generate sample points representing load forecast error, wind power forecast error, and photovoltaic forecast error. These sample points are used to simulate different error scenarios.

[0191] S24: Calculate the net load error of the sample points and divide it into positive error samples and negative error samples.

[0192] In order to estimate the net load forecasting error (NLF), the calculation formula of the net load forecasting error NE is first defined as:

[0193]

[0194] in, represents the load forecast value, represents the wind power forecast value, Represents the photovoltaic prediction value. The distribution of reflects the cumulative error effect under different prediction scenarios. In order to expand to more new energy scenarios, the wind power prediction value and photovoltaic prediction value can be the sum of the prediction values ​​of more power generation projects M.

[0195] In order to estimate the uncertainty range of the positive and negative prediction errors NF, the NF samples are divided into positive errors (NF > 0) and negative errors (NF < 0), and their standard deviations σ are calculated respectively.

[0196] S25: Calculate the upper and lower standard deviations of the net load error based on the positive error samples and the negative error samples. The calculation expressions of the upper and lower standard deviations are as follows:

[0197]

[0198]

[0199] Where, and They are the upper standard deviation and the lower standard deviation, which are used to indicate the degree of fluctuation of the load data prediction error; and are the number of positive error samples and negative error samples, respectively. and They are the i-th net load error sample in the positive error sample and the j-th net load error sample in the negative error sample respectively.

[0200] The above formula can be used to obtain the upper and lower fluctuation range of the net load forecast error, which is used to describe the uncertainty of the forecast.

[0201] S26: Correct the regional high-resolution load data according to the upper and lower bound standard deviations as follows;

[0202]

[0203]

[0204] Where, is the load data after correction at time t, is the load data at time t in the regional high-resolution load data, is the upper bound of the load forecast error, is the lower bound of the load forecast error.

[0205] The principle of this embodiment is to use the Gumble-copula joint probability distribution function, combined with Monte Carlo sampling, to correct load curve forecast errors. By assuming that the forecast errors for load, wind power, and photovoltaic power generation follow normal, Weibull, and Beta distributions, respectively, the correlation between them is modeled using a copula function. Combined with Monte Carlo sampling, this method can simulate forecast errors in different scenarios and their impact on the overall net load error. These distribution assumptions lay the theoretical foundation for subsequent error analysis and uncertainty assessment.

[0206] In this way, the load forecast value can be effectively corrected, taking into account the uncertainty impact of wind power, photovoltaic and other renewable energy forecast errors on load forecasting, thereby providing a more accurate load forecast range for future power grid scheduling.

[0207] In one embodiment, Monte Carlo sampling is used to generate sample points from the joint distribution. Through Monte Carlo sampling, multiple random samples can be generated based on the marginal distribution, and these samples are associated through the Copula function to generate load and renewable energy samples with dependencies. This includes:

[0208] S231: Sample the variables for each prediction item from a uniform distribution.

[0209] From the standard uniform distribution For each variable (such as load , wind power and photovoltaic ) Generate random numbers:

[0210]

[0211] These sampling points Will be passed as input to the Copula function.

[0212] S232: The sampled uniformly distributed samples are inversely transformed by the Gumbel Copula function to generate samples with dependency.

[0213] The uniformly distributed samples will be sampled Generate samples with dependencies through the inverse Copula transform. For Gumbel Copula, the specific formula for the inverse transform is:

[0214]

[0215] Similarly, we can get and Through this process, correlated samples that conform to the joint distribution are obtained.

[0216] S233: Map the generated Copula samples back to the original variable space through the inverse cumulative distribution function of the marginal distribution to obtain the actual error samples of each predicted item, where the error samples are sample points.

[0217] Inverse cumulative distribution function via marginal distribution , the generated Copula sample Mapping back to the original variable space, we can obtain the actual error values ​​of load, wind power, and photovoltaic power generation:

[0218]

[0219] in, is the inverse cumulative distribution function of load, wind power, and photovoltaic power generation. In this way, the generated samples not only follow the marginal distribution but also reflect the correlation between them.

[0220] In the generation of multiple loads , wind power and photovoltaic After , the net load error at each sampling point can be calculated:

[0221]

[0222] For each sample , which is divided into the forward error and negative error .

[0223] a) Divide the net load error samples into positive and negative errors

[0224] A positive error indicates that the predicted load is higher than the actual situation, and a negative error indicates that the predicted load is lower than the actual situation. The positive and negative error samples are defined as:

[0225]

[0226] b) Calculate the standard deviation of positive and negative errors

[0227] For positive and negative error samples, their standard deviations σt(NF > 0) and σt(NF < 0) are calculated respectively to estimate the upper and lower bound uncertainties of the net load forecast.

[0228] In one embodiment, the rolling average method is used to calculate the regional frequency modulation demand forecast at each sampling moment. Specifically, the rolling average method is used to calculate the regional frequency modulation demand forecast. and The rate of change is two values. First, define: is the regional frequency modulation demand at time t. is the regional control deviation at time t, For all units Capacity, this method does not involve the allocation of units , regional control deviation Value equal to (Before considering various correction factors, detailed corrections are given later).

[0229] The implementation process of this step includes:

[0230] S31: Record the load data of each load point in the corrected regional high-resolution load data as first load data.

[0231] S32: For each load point, take several load values ​​before and after and perform sliding average to obtain load data of each load point processed by the rolling average method, and record it as the second load data.

[0232] The rolling average method uses a sliding window to average the load values, thereby obtaining a smooth load curve. If the load sampling period is set to 15 minutes, K load values ​​are taken before and after each load point for sliding average. The sliding average calculation formula is as follows:

[0233]

[0234] in, is the time after smoothing by the sliding average method The load value, that is, the second load data, It is the original time The load value, i.e., the first load data; The value of K directly affects the smoothness of the load curve. Value, sampling period 15 seconds .

[0235] S33: The difference between the first load data and the second load data of each load point is used as an estimate of the secondary frequency regulation capacity required by the control area at the corresponding moment.

[0236] The secondary frequency regulation capacity required in the control area at time t is estimated to be

[0237]

[0238] In this amendment, By considering the volatility of new energy forecast errors, the load forecast is made more accurate. yes Secondary frequency modulation capacity required in the time control area, yes Regional control deviation of the time control area, Indicates that the control area needs to increase capacity. Indicates that the control area needs to reduce capacity.

[0239] S34: The difference between the first load data of two adjacent load points is used as the required load of the control area at the corresponding time. .

[0240] To further calculate Regulation rate, quantified by the rate of load change at adjacent moments Adjust the rate. Definition The regulation rate is the difference in load changes between adjacent moments:

[0241]

[0242] in, yes The rate of change between adjacent load values ​​at a given moment.

[0243] In one embodiment, the regional frequency regulation demand forecast is modified using a control criterion for power system dispatch, including:

[0244] S41: Determine the regional control deviation at time t according to the BAAL criterion The upper and lower limit conditions met

[0245] In order to meet the BAAL criterion, the regional control deviation ACEt of the control area at a certain moment must meet the following conditions:

[0246]

[0247]

[0248] in, and It is based on the upper and lower limits of frequency deviation and is calculated by the following formula:

[0249]

[0250]

[0251] Where, B is the frequency deviation coefficient of the control area, and are the upper and lower reference frequencies, To correct frequency deviation. The calculation formula is as follows:

[0252]

[0253] Where, is the marginal change in power, is the frequency response coefficient of the system;

[0254] S42: Control deviation by region and frequency deviation Correction of secondary frequency regulation capacity estimate.

[0255] Considering regional control deviations in the control area and frequency deviation , we can get the control area at time The required secondary frequency regulation capacity requirements are estimated to be:

[0256]

[0257] Where, It is the secondary frequency regulation capacity requirement of the control area determined by considering the BAAL standard, in MW.

[0258] In one embodiment, selecting the optimal standard deviation value as the demand forecast value for power system scheduling in each scheduling period based on statistical principles includes:

[0259] S51: Take is the optimal standard deviation value; is the standard deviation of the regional frequency regulation demand forecast.

[0260] S52: Calculate the optimal standard deviation value for each scheduling period in the day in sequence as the demand forecast value for power system scheduling, and obtain the time-based forecast result of the secondary frequency regulation demand in the day.

[0261] According to statistics, approximately 99.8% of samples from a normal distribution fall within the interval (-3σ, 3σ). Since the frequency regulation capacity and rate demands derived from the aforementioned calculation method are calculated every 15 seconds, and the minimum time accuracy for power system dispatch is 15 minutes, all capacity and rate demand forecasts within each 15-minute period are selected, their σ values ​​are calculated, and 3σ is taken as the capacity and rate demand values ​​for that period. The capacity and rate demand values ​​are then calculated sequentially for 96 points in time during the day.

[0262] Specifically, the probability density function of the normal distribution is:

[0263]

[0264] in, is the mean, is the standard deviation. Coverage probability within the interval , so it can be considered that The demand capacity within this range can meet 99.8% of frequency regulation needs. This approach allows for reasonable prediction and determination of AGC demand every 15 minutes, ensuring the stable operation of the intraday dynamic frequency regulation market.

[0265] Based on the same inventive concept, an embodiment of the present application further provides a device for predicting the regional secondary frequency regulation demand of a power system by time period, for implementing the aforementioned method for predicting the regional secondary frequency regulation demand of a power system. The implementation solution provided by this device is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations in the embodiment of the device for predicting the regional secondary frequency regulation demand of a power system by time period provided below can be found in the limitations of the method for predicting the regional secondary frequency regulation demand of a power system by time period, and will not be repeated here.

[0266] See also Figure 4 The embodiment of the present invention further provides a time-division prediction device for regional secondary frequency regulation demand of a power system, comprising:

[0267] The data acquisition and processing module is used to collect regional load data and predict regional high-resolution load data through a pre-trained super-resolution deep learning model;

[0268] A first correction module is used to determine the degree of fluctuation of the load data prediction error by utilizing the uncertainty of the new energy power generation prediction, and to correct the regional high-resolution load data based on the degree of fluctuation of the prediction error;

[0269] The first prediction module is used to calculate the regional frequency regulation demand forecast at each sampling moment using a rolling average method based on the corrected regional high-resolution load data;

[0270] A second correction module is used to correct the regional frequency regulation demand forecast using the control criterion of power system dispatch;

[0271] The second prediction module is used to calculate the standard deviation of the regional frequency regulation demand forecast at all sampling moments in each scheduling period by time period, and select the best standard deviation value as the demand forecast value of the power system scheduling in each scheduling period based on statistical principles. The best standard deviation value is the standard deviation value covering the set proportional scheduling demand in the scheduling period.

[0272] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0273] Reference Figure 5 An embodiment of the present invention also provides a computer device, including: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, it implements the time-sharing prediction method for the regional secondary frequency regulation demand of the power system as described in any one of the above methods.

[0274] The computer device may be a desktop computer, notebook computer, PDA, cloud server or other computing device. The computer device may include, but is not limited to, a processor and a memory. It will be understood by those skilled in the art that Figure 5 The computer device is merely an example and does not constitute a limitation on the computer device. The computer device may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, etc.

[0275] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0276] In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard drive or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped with the computer device. Furthermore, the memory may include both an internal storage unit of the computer device and an external storage device. The memory is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory may also be used to temporarily store data that has been output or is about to be output.

[0277] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting the regional secondary frequency regulation demand of a power system by time period as described in any one of the above methods is implemented.

[0278] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a camera / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0279] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0280] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0281] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0282] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A time-division forecasting method for regional secondary frequency regulation demand in a power system, characterized in that: The steps include: Collect regional load data and predict regional high-resolution load data using a pre-trained super-resolution deep learning model; Determining the degree of fluctuation of load data prediction errors using the uncertainty of renewable energy power generation predictions, and correcting the high-resolution load data of the region based on the degree of fluctuation of the prediction errors; Based on the revised regional high-resolution load data, a rolling average method is used to calculate the regional frequency regulation demand forecast at each sampling moment; Correcting the regional frequency regulation demand forecast using a control criterion for power system dispatch; The standard deviation of the regional frequency regulation demand forecast at all sampling moments in each scheduling period is calculated by time period, and the optimal standard deviation value is selected as the demand forecast value of the power system scheduling in each scheduling period based on statistical principles. The optimal standard deviation value is the standard deviation value covering the set proportional scheduling demand in the scheduling period.

2. The method for predicting regional secondary frequency regulation demand of a power system according to claim 1, characterized in that: The super-resolution deep learning model adopts a deep learning network model based on an encoder-decoder structure. The deep learning network model uses the Adam optimizer for model training. The training process is performed by minimizing a predefined total loss function. The total loss function is as follows: Where, is the total loss function; is the reconstruction error function, For real high-resolution data, High-resolution data output for the model; is the regularization term, is the regularization coefficient.

3. The method for predicting regional secondary frequency regulation demand of a power system according to claim 1, characterized in that: Determining the degree of fluctuation of load data prediction error using the uncertainty of renewable energy power generation prediction, and correcting the regional high-resolution load data based on the degree of fluctuation of the prediction error, including: According to the uncertainty of forecasting various items in the new energy power generation forecast, the error distribution of various forecast items is determined; For the error distribution of various forecast items, the Gumbel Copula function is used to establish a joint distribution; Generating sample points of each type of the predicted items from the joint distribution; The net load error of the sample points is calculated and divided into positive error samples and negative error samples; the calculation expression of the net load error is as follows: Where, is the net load error, when it is greater than zero, the corresponding net load error sample is classified as the positive error sample, and when it is less than zero, the corresponding net load error sample is classified as the negative error sample; The sample point predicted value of the load in the forecast project; The sum of the predicted values ​​of the sample points of the power generation projects in the forecast project; The upper and lower standard deviations of the net load error are calculated based on the positive error samples and the negative error samples. The upper and lower standard deviations are calculated as follows: Where, and are the upper standard deviation and the lower standard deviation, respectively, which are used to indicate the degree of fluctuation of the load data prediction error; and are the number of the positive error samples and the number of the negative error samples, respectively, and are respectively the i-th net load error sample in the positive error sample and the j-th net load error sample in the negative error sample; The high-resolution load data of the region is corrected according to the upper and lower standard deviations as follows; Where, is the load data after correction at time t, is the load data at time t in the high-resolution load data of the region, is the upper bound of the load forecast error, is the lower bound of the load forecast error.

4. The method for predicting regional secondary frequency regulation demand of a power system according to claim 3, characterized in that: Generating sample points from the joint distribution includes: Sampling a variable for each of the predicted items from a uniform distribution; The uniformly distributed samples are transformed by the inverse Gumbel Copula function to generate samples with dependency; The generated Copula samples are mapped back to the original variable space through the inverse cumulative distribution function of the marginal distribution to obtain the actual error samples of each predicted item, where the error samples are the sample points.

5. The method for predicting regional secondary frequency regulation demand of power system according to claim 1, characterized in that: The rolling average method is used to calculate the regional frequency regulation demand forecast at each sampling moment, including: The load data of each load point in the corrected regional high-resolution load data is recorded as first load data; For each load point, a plurality of load values ​​before and after are taken and sliding averaged to obtain the load data of each load point processed by the rolling average method, and record it as the second load data; taking the difference between the first load data and the second load data of each load point as an estimate of the secondary frequency regulation capacity required by the control area at the corresponding moment; The difference between the first load data of two adjacent load points is used as the required load of the control area at the corresponding time. .

6. The method for predicting regional secondary frequency regulation demand of a power system according to claim 1, characterized in that: The regional frequency regulation demand forecast is modified using a control criterion for power system dispatch, including: According to the BAAL criterion, determine the regional control deviation at time t The upper and lower limit conditions are met; the upper and lower limit values ​​are determined according to the following formula: Where, and Regional control deviation The upper and lower limits of B are the frequency deviation coefficient of the control area. and are the upper and lower reference frequencies, To correct the frequency deviation, is the frequency deviation at time t; the calculation formula for correcting the frequency deviation is as follows: Where, is the marginal change in power, is the frequency response coefficient of the system; Control deviation according to the area and frequency deviation The secondary frequency regulation capacity estimate is revised as follows: Where, It is the estimated value of secondary frequency regulation capacity demand in the control area determined by considering the BAAL criterion.

7. The method for predicting regional secondary frequency regulation demand of a power system according to claim 1, characterized in that: Selecting the best standard deviation value based on statistical principles as the demand forecast value for power system dispatch in each dispatch period includes: Pick is the optimal standard deviation value; is the standard deviation of the frequency regulation demand forecast for the region; The optimal standard deviation value of each scheduling period within the day is calculated in sequence as the demand forecast value of power system scheduling, and the time period forecast result of the secondary frequency regulation demand within the day is obtained.

8. A time-division forecasting device for regional secondary frequency regulation demand in a power system, characterized in that: include: The data acquisition and processing module is used to collect regional load data and predict regional high-resolution load data through a pre-trained super-resolution deep learning model; a first correction module, configured to determine a degree of fluctuation of a load data prediction error by utilizing uncertainty in the prediction of renewable energy power generation, and to correct the regional high-resolution load data based on the degree of fluctuation of the prediction error; A first prediction module is configured to calculate a regional frequency regulation demand forecast at each sampling moment using a rolling average method based on the corrected regional high-resolution load data; A second correction module is used to correct the regional frequency regulation demand forecast using a control criterion for power system dispatching; The second prediction module is used to calculate the standard deviation of the regional frequency regulation demand forecast at all sampling moments in each scheduling period by time period, and select the optimal standard deviation value as the demand forecast value of the power system scheduling in each scheduling period based on statistical principles. The optimal standard deviation value is the standard deviation value covering the set proportional scheduling demand in the scheduling period.

9. A computer device, characterized in that: The device includes a processor and a memory: The memory is used to store the computer program and send instructions of the computer program to the processor; The processor executes the time-division prediction method for regional secondary frequency regulation demand of a power system according to any one of claims 1 to 7 according to the instructions of the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for predicting regional secondary frequency regulation demand of a power system by time period is implemented according to any one of claims 1 to 7.

Images