Power grid load prediction method and system based on time division

Abstract

The invention discloses a power grid load prediction method and system based on time division, and relates to the technical field of power grids, and the method comprises the steps: mining sub-region data after time division, and carrying out the analysis of the sub-region data; dividing the decomposed sub-region data by using an ensemble learning algorithm, and combining a decomposition result and a division result by using a time sequence prediction model; and collecting a power grid load coefficient during a historical shutdown period, and predicting the power grid load based on the automatically filled data. According to the method, the defects of a traditional prediction method in processing multi-dimensional factors and data missing are effectively overcome, the load air temperature is accurately estimated, the prediction precision and robustness are improved, the continuity and accuracy of time-phased load prediction are ensured by prolonging automatic filling of a load curve axis according to special conditions such as a shutdown period, and the load prediction efficiency is improved. And the reliability and efficiency of operation management of the power system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and in particular to a method and system for predicting power grid load by time periods. Background Art

[0002] The rapid development of today's society and the continuous growth of energy demand have put forward higher requirements for the operation and management of power systems. With the popularization of smart grids and the diversification of energy structures, the dynamic changes of power grid loads have become more complex and difficult to predict, making accurate power grid load forecasting a key link in power dispatching and optimal resource allocation, which is of great significance to improving the stability and reliability of power grid operation.

[0003] At present, in the existing technology, the power grid load forecasting method often relies on a single model or simple statistical analysis, lacks comprehensive consideration of multi-dimensional factors, and has low prediction accuracy. In addition, the existing methods are difficult to deal with data missing problems in special circumstances such as shutdown periods, resulting in inaccurate load forecasting. The multi-model integration and automatic filling technology of the present invention can effectively solve the above-mentioned defects and improve the accuracy and robustness of power grid load forecasting. Summary of the invention

[0004] In view of the problems existing in the existing time-divided power grid load forecasting method and system, the present invention is proposed.

[0005] Therefore, the purpose of the present invention is to provide a method and system for predicting power grid load by time period. The present invention solves the problem of data missing in special circumstances such as shutdown periods by using a method for predicting power grid load by time period.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting power grid load by time period division, which includes mining sub-region data after time period division, analyzing the sub-region data, and using a combined load prediction algorithm to decompose the analyzed sub-region data;

[0008] The decomposed sub-region data are divided into components using an integrated learning algorithm, and the decomposition results and component results are combined using a time series prediction model;

[0009] Collect the grid load factor during historical shutdown periods, predict the extended load curve axis of the grid load factor, automatically fill in the regional data divided into time periods in this stage according to the extended load curve axis, and predict the grid load based on the automatically filled data.

[0010] As a preferred solution of the power grid load prediction method divided by time periods of the present invention, wherein: the mining of the sub-region data after the time period division includes calculating the overall power grid load factor of the sub-region data according to the improved Spearman correlation coefficient, the overall power grid load factor includes calculating the load temperature of the power grid load, and optimizing the overall power grid load factor based on the load temperature;

[0011] The specific calculation formula of the load air temperature is:

[0012]

[0013] Among them, xi represents the candidate characteristic value of the overall power grid load factor, represents the mean value of the candidate feature value, yi represents the load value divided in this time period, represents the average value of load value, r represents the load temperature;

[0014] When the absolute value |r|→1, the attribute correlation of the load temperature is enhanced;

[0015] When the absolute value |r|→0, the attribute correlation of the load temperature is weakened.

[0016] As a preferred solution of the time-divided power grid load prediction method of the present invention, wherein: the analysis of the sub-region data includes estimating the strong load temperature and the weak load temperature according to the statistical bootstrap method, the estimation includes repeatedly sampling the overall power grid load factor to generate multiple sampling samples, and obtaining the statistics of the load temperature according to the generated multiple sampling samples;

[0017] Decomposing the analyzed sub-region data includes dynamically programming the statistics of the load temperature and defining the Ferecher distance, wherein the definition of the Ferecher distance includes defining it according to the difference of the load temperature counted on the day;

[0018] The specific calculation formula for the repeated sampling is:

[0019]

[0020] Where n is the sample size, Indicates the high load temperature, Indicates the weak load temperature, represents the strong load temperature of the i-th observation in the b-th resampled sample, represents the weak load temperature of the i-th observation in the b-th resampled sample, represents the standard deviation of the strong load temperature in the bth resampled sample, represents the standard deviation of the weak load air temperature in the bth resampled sample.

[0021] As a preferred solution of the time-divided power grid load forecasting method of the present invention, wherein: the decomposition of the analyzed sub-area data includes decomposing the statistics based on the Fourier decomposition algorithm, and the decomposition result of the statistics includes a daily cycle component, a cycle component, a low-frequency component, a high-frequency component and a constant term. The specific calculation formula of the decomposition is:

[0022] y(t)=g(t)+s(t)+h(t)+ε t

[0023] Among them, y(t) represents the statistical result data after decomposition, g(t) represents the trend component, s(t) represents the periodic component, h(t) represents the date impact component, ε t represents the residual term, t represents the time point;

[0024] The low-frequency component is used as the temperature sensitive component, and the temperature sensitive component includes the SVM algorithm, the Light GBM algorithm and the Ridge algorithm, which are weighted in a ratio of 3:6:1, and the integrated learning model is input with the forecast temperature of the day to obtain the prediction result;

[0025] The periodic component, high-frequency component and constant term are added together as the time series component, and the time series load forecasting is completed based on the Prophet algorithm;

[0026] The specific calculation formula of the SVM algorithm is:

[0027]

[0028] Among them, f(x) represents the regression prediction output value of the SVM algorithm, α i represents the Lagrange multiplier, K(x i , x) represents the kernel function, x i represents the candidate characteristic value of the overall power grid load factor, x represents the input characteristic vector to be predicted, and b represents the resampled sample;

[0029] The specific calculation formula of the Light GBM algorithm is:

[0030]

[0031] Among them, F(x) represents the regression prediction output value of the Light GBM algorithm, γ m represents the weight coefficient of the mth weak learner, T m (x) represents the output of the mth decision tree, and M represents the total number of weak learners;

[0032] The specific calculation formula of the Ridge algorithm is:

[0033]

[0034] Among them, J(θ) represents the regression prediction output value of the Ridge algorithm, yi represents the load value divided in this time period, θ represents the regression coefficient vector, and θ T x i represents the predicted value of the regression model, λ represents the regularization parameter, represents the square of the regression coefficient;

[0035] The specific calculation formula of the Prophet algorithm is:

[0036]

[0037] in, The table shows the regression prediction output value of the Prophet algorithm, and y(t) represents the statistical result data after decomposition.

[0038] As a preferred solution of the method for predicting power grid load by time period division of the present invention, wherein: the merging of the decomposition result and the component result includes connecting the decomposed sub-region data into a curve for smoothing, horizontally stretching the curve according to the stage divided at this time, performing time period division twice, and constructing the load factor of the stage divided at this time according to the daily change ratio of the new and old coefficients;

[0039] If the load factor of the first time period division is greater than the load factor of the second time period division, the future predicted load factor will be corrected according to the average error rate of the previous day's time period division.

[0040] Otherwise, all predicted load factors are aggregated to obtain the total predicted load before and after the stage divided at this time.

[0041] As a preferred solution of the time-divided power grid load prediction method of the present invention, wherein: the extended load curve axis of the predicted power grid load factor includes pre-processing the lateral stretching, and selecting the precision of the time-divided division according to the stretching ratio;

[0042] If there are load peaks or troughs during the historical downtime period, the stretch ratio is determined by using a smoothing method of local expansion or contraction;

[0043] If there is no load change feature, the default stretch ratio is used; if there is a trough during the historical downtime period, the time series during the historical downtime period is analyzed to identify the long-term trend and seasonal changes of the time series.

[0044] As a preferred solution of the time-divided power grid load prediction method of the present invention, wherein: the automatic filling of the regional data divided by time periods in the current stage according to the extended load curve axis includes mapping the extended load curve axis with the regional data divided by time periods at the current stage;

[0045] The data of the extended load curve axis is mapped to the regional data divided into the current time periods. By comparing the sequence shapes of the two during the mapping process, the matching path is found to complete the power grid load forecast.

[0046] In the second aspect, an embodiment of the present invention provides a power grid load forecasting system divided into time periods, which includes: an analysis module, which mines sub-regional data after the time period division and analyzes the sub-regional data; a prediction module, which uses an integrated learning algorithm to componentize the decomposed sub-regional data, and adopts a time series prediction model to merge the decomposition result and the component result; a filling module, which collects the power grid load factor during the historical shutdown period, automatically fills the regional data divided into time periods in this stage according to the extended load curve axis, and predicts the power grid load based on the automatically filled data.

[0047] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, it implements any step of the above-mentioned time-divided power grid load forecasting method.

[0048] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the above-mentioned method for predicting power grid load by time period is implemented.

[0049] The beneficial effects of the present invention are as follows: the present invention effectively solves the shortcomings of traditional prediction methods in dealing with multi-dimensional factors and data missing problems through multi-model integration algorithms and automatic filling technology, and uses the improved Spearman correlation coefficient and statistical bootstrap method to accurately estimate the load temperature, and improves the accuracy and robustness of the prediction based on the time series prediction model. In addition, for special circumstances such as shutdown periods, the automatic filling of the load curve axis is extended to ensure the continuity and accuracy of time-divided load forecasting, thereby improving the reliability and efficiency of power system operation management. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. 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 creative work. Among them:

[0051] Figure 1 A specific flow chart of a method and system for predicting power grid load by time period provided in one embodiment of the present invention.

[0052] Figure 2 A schematic diagram of a prediction algorithm of a method and system for predicting power grid load by time periods provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0056] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0057] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0058] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0059] Example 1

[0060] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a method for predicting power grid load by time period division, comprising:

[0061] S1: Mining the sub-regional data after time division, analyzing the sub-regional data, and using the combined load forecasting algorithm to decompose the analyzed sub-regional data.

[0062] Among them, mining the sub-regional data after time division includes calculating the overall power grid load factor of the sub-regional data according to the improved Spearman correlation coefficient, the overall power grid load factor includes calculating the load temperature of the power grid load, and optimizing the overall power grid load factor based on the load temperature;

[0063] The specific calculation formula of load temperature is:

[0064]

[0065] Among them, xi represents the candidate characteristic value of the overall power grid load factor, represents the mean value of the candidate feature value, yi represents the load value divided in this time period, represents the average value of load value, r represents the load temperature;

[0066] When the absolute value |r|→1, the attribute correlation of the load temperature is enhanced;

[0067] When the absolute value |r|→0, the attribute correlation of the load temperature is weakened.

[0068] Furthermore, by mining the sub-regional data after time division, the load characteristics of each sub-region are analyzed. Specifically, the improved Spearman correlation coefficient is used to calculate the overall grid load factor of each sub-region. In order to improve the prediction accuracy, the overall grid load factor in the sub-regional data is first calculated. This coefficient represents the sensitivity of the grid load to the temperature. According to the statistical results, the candidate feature values ​​of the load temperature are determined, and the mean of these candidate features is calculated.

[0069] The candidate eigenvalues ​​may be historical load data in different seasons or under special weather conditions. The load temperature may be calculated by comparing the mean of these data with the actual load value in a specific period of time. If the absolute difference between the calculated load temperature and a certain eigenvalue is close to 1, it indicates that the load temperature is highly sensitive to specific factors, such as extreme weather; if it is close to 0, it indicates that the sensitivity is relatively weak. In this way, the present invention can optimize the overall power grid load factor and make the prediction model more accurate and robust. In practical applications, this method can significantly improve the operating efficiency of the power grid under different load conditions, especially in load management in response to extreme weather or special events.

[0070] S1.1: Analyzing the sub-regional data includes estimating the strong load temperature and the weak load temperature according to the statistical bootstrap method, the estimation includes repeatedly sampling the overall grid load factor to generate multiple sampling samples, and obtaining the statistics of the load temperature according to the generated multiple sampling samples;

[0071] Decomposing the analyzed sub-area data includes dynamically programming the statistics of the load and temperature, defining the Ferecher distance, and defining the Ferecher distance includes defining it according to the difference of the load and temperature statistics of the day;

[0072] The specific calculation formula for repeated sampling is:

[0073]

[0074] Where n is the sample size, Indicates the high load temperature, Indicates the weak load temperature, represents the strong load temperature of the i-th observation in the b-th resampled sample, represents the weak load temperature of the i-th observation in the b-th resampled sample, represents the standard deviation of the strong load temperature in the bth resampled sample, represents the standard deviation of the weak load air temperature in the bth resampled sample.

[0075] Furthermore, in the power grid load forecasting method of the present invention, the analysis of the sub-region data first includes using the statistical bootstrap method to estimate the strong load temperature and the weak load temperature, and generating multiple sampling samples by repeated sampling of the overall power grid load factor to obtain the statistics of the load temperature. For example, in a certain sub-region, assuming that the sample size is 1000 times, the result of each sampling will produce a different load temperature value. By counting the load temperature values ​​in these samples, the standard deviation and distribution of the strong load temperature and the weak load temperature can be calculated respectively.

[0076] The analyzed sub-region data is decomposed, and the statistics of the load temperature are calculated using a dynamic programming method. The change of the load temperature is quantified by defining the Ferecher distance, which measures the difference between the strong load temperature and the weak load temperature. Specifically, based on the statistical data of the day, the difference in load temperature can be calculated, and these differences are used to define the Ferecher distance. This method can capture the load change law of the power grid under different temperature conditions, especially in extreme weather conditions, which helps to more accurately predict and manage the load changes of the power grid. Through these detailed data processing and analysis, the present invention can provide more accurate power grid load prediction results, thereby improving the operating efficiency and stability of the power grid.

[0077] S1.2: Decomposing the analyzed sub-region data includes decomposing the statistics based on the Fourier decomposition algorithm. The decomposition results of the statistics include daily cycle components, cycle components, low-frequency components, high-frequency components and constant terms. The specific calculation formula for the decomposition is:

[0078] y(t)=g(t)+s(t)+h(t)+ε t

[0079] Among them, y(t) represents the statistical result data after decomposition, g(t) represents the trend component, s(t) represents the periodic component, h(t) represents the date impact component, ε t represents the residual term, t represents the time point;

[0080] The low-frequency component is used as the temperature sensitive component. The temperature sensitive component includes the SVM algorithm, the Light GBM algorithm and the Ridge algorithm. The weighting ratio is 3:6:1. The integrated learning model is input with the forecast temperature of the day to obtain the forecast result.

[0081] The periodic component, high-frequency component and constant term are added together as the time series component, and the time series load forecasting is completed based on the Prophet algorithm;

[0082] The specific calculation formula of the SVM algorithm is:

[0083]

[0084] Among them, f(x) represents the regression prediction output value of the SVM algorithm, α i represents the Lagrange multiplier, K(x i , x) represents the kernel function, x i represents the candidate characteristic value of the overall power grid load factor, x represents the input characteristic vector to be predicted, and b represents the resampled sample;

[0085] The specific calculation formula of the Light GBM algorithm is:

[0086]

[0087] Among them, F(x) represents the regression prediction output value of the Light GBM algorithm, γ m represents the weight coefficient of the mth weak learner, T m (x) represents the output of the mth decision tree, and M represents the total number of weak learners;

[0088] The specific calculation formula of the Ridge algorithm is:

[0089]

[0090] Among them, J(θ) represents the regression prediction output value of the Ridge algorithm, yi represents the load value divided in this time period, θ represents the regression coefficient vector, and θ T x i represents the predicted value of the regression model, λ represents the regularization parameter, represents the square of the regression coefficient;

[0091] The specific calculation formula of the Prophet algorithm is:

[0092]

[0093] in, The table shows the regression prediction output value of the Prophet algorithm, and y(t) represents the statistical result data after decomposition.

[0094] In the power grid load forecasting method of the present invention, the analyzed sub-region data is firstly subjected to Fourier decomposition to obtain daily cycle components, periodic components, low-frequency components, high-frequency components and constant terms. For example, the daily cycle component of a certain area shows that the load reaches 2500MW during the peak hours in the morning and evening in summer, and 2000MW in winter. The periodic component captures the load drop every weekend, from 1500MW to 1200MW, and seasonal changes such as summer load of 1800MW and winter load of 1600MW are also included. The low-frequency component reflects the long-term load growth trend, from 1000MW to 1300MW, and the high-frequency component records the changes in short-term load caused by sudden cold waves from 1500MW to 1700MW, and the constant term remains at a stable level of 1000MW.

[0095] The low-frequency component is used as the temperature-sensitive component, and is predicted by a weighted combination of SVM, Light GBM and Ridge algorithms. Assuming that the temperature drops from 20°C to 10°C, resulting in an increase in load from 1200MW to 1400MW, the prediction results of these three algorithms can accurately predict the increase in load under cooling conditions. The synthetic results of the periodic component, high-frequency component and constant term are used to predict time series through the Prophet algorithm. Assuming that the load drops from 1500MW to 800MW during the Spring Festival, the Prophet algorithm can accurately predict the load changes during future holidays. Through this comprehensive analysis and prediction, the present invention can improve the accuracy of power grid load prediction and provide a scientific basis for power grid scheduling and planning. The component prediction result data is shown in Table 1 below:

[0096] Table 1 Component prediction result data

[0097]

[0098] The table shows the main data involved in load forecasting. First, the data includes the decomposition results of the power grid load in different time periods, including daily cycle components, cycle components, low-frequency components, high-frequency components and constant terms. These data provide a detailed perspective on load fluctuations and reflect periodic changes and trends. The table also shows the forecast results based on the ensemble learning model, including the forecast outputs of SVM, Light GBM and Ridge algorithms. The forecast results of each algorithm are compared in the actual power grid load. Combined with the forecast results of the time series forecasting model (Prophet algorithm), they are used together to build a comprehensive power grid load forecasting picture. Through these data, we can better understand the law of load changes and make more accurate power grid load forecasts.

[0099] S2: Use the ensemble learning algorithm to divide the decomposed sub-region data into components, and use the time series prediction model to merge the decomposition results and the component results.

[0100] The merging of the decomposition results and the component results includes connecting the decomposed sub-area data into a curve for smoothing, horizontally stretching the curve according to the stage divided at this time, and dividing it into time periods for the second time, and constructing the load coefficient of the stage divided at this time according to the daily change ratio of the new and old coefficients;

[0101] If the load factor of the first time period division is greater than the load factor of the second time period division, the future predicted load factor will be corrected according to the average error rate of the previous day's time period division.

[0102] Otherwise, all predicted load factors are aggregated to obtain the total predicted load before and after the stage divided at this time.

[0103] Furthermore, in the process of using an integrated learning algorithm to componentize the decomposed sub-regional data and merge it with the time series prediction model, the decomposed sub-regional data are first connected into smooth curves to better capture the changing trends in the data. These curves are stretched horizontally according to the current stage to adjust the time dimension of the data to adapt it to different time period divisions.

[0104] Assume that in the power grid load data of a certain stage, the decomposed periodic components and low-frequency components are connected into a smooth curve. If the load factor obtained by the first time period division is 0.75, and the load factor obtained by the second time period division is 0.70, it is necessary to make corrections based on the average error rate of the previous day's time period division. If the average error rate of the previous day is 10%, the future predicted load factor will be adjusted. If the load factor of the first division is greater than the second, the load forecast will be adjusted by subtracting the average error rate; otherwise, all predicted load factors are summarized to finally obtain the power grid load forecast value of this stage.

[0105] If the load factors shown in the time period division results are 0.75 (first time) and 0.70 (second time), and the average error rate of the previous day is 10%, the adjusted load forecast factor will be closer to the actual load, thereby improving the accuracy of the forecast. This method effectively integrates various data of load forecasts at each stage, ensuring the comprehensiveness and accuracy of load forecasts.

[0106] S3: Collect the grid load factor during historical shutdown periods, predict the extended load curve axis of the grid load factor, automatically fill in the regional data divided by time periods in this stage according to the extended load curve axis, and predict the grid load based on the automatically filled data.

[0107] Among them, the extended load curve axis for predicting the load factor of the power grid includes preprocessing the horizontal stretching and selecting the accuracy of the time period division according to the stretching ratio;

[0108] If there are load peaks or troughs during the historical downtime period, the stretch ratio is determined by using a smoothing method of local expansion or contraction;

[0109] If there is no load change feature, the default stretch ratio is used; if there is a trough during the historical downtime, the time series during the historical downtime is analyzed to identify the long-term trend and seasonal changes of the time series.

[0110] Furthermore, after collecting the grid load factor during historical shutdown periods, it is first necessary to predict the extended load curve axis of the grid load factor. To this end, the lateral stretching needs to be preprocessed to adjust the time dimension of the data and select the appropriate time period division accuracy.

[0111] For historical data with significant load peaks or troughs, a local expansion or reduction smoothing method is used to determine the stretching ratio. For example, if during a historical shutdown period, the power grid load data peaks from 10 a.m. to 12 a.m. and troughs from 2 a.m. to 4 a.m., then expansion smoothing will be performed in the peak area and reduction smoothing will be performed in the trough area to more accurately capture these load changes. If the data does not show obvious load change characteristics, the default stretching ratio is used, such as a 1.2 times time expansion ratio.

[0112] When there are obvious troughs during historical shutdown periods, it is necessary to conduct in-depth analysis of the time series to identify long-term trends and seasonal changes. For example, if it is found that the grid load is generally low in the summer during multiple shutdown cycles, the load factor will be adjusted in combination with these seasonal changes to reflect seasonal load changes. The extended load curve axis data will be mapped and automatically filled with the regional data divided by time periods in the current stage to ensure that all data points are in line with the expected load trend. Based on these automatically filled data, the grid load can be accurately predicted, which improves the reliability and accuracy of the prediction.

[0113] S3.1: Automatically filling the regional data divided by time periods in this stage according to the extended load curve axis includes mapping the extended load curve axis with the regional data divided by current time periods;

[0114] The data of the extended load curve axis is mapped to the regional data divided into the current time periods. By comparing the sequence shapes of the two during the mapping process, the matching path is found to complete the power grid load forecast.

[0115] Preferably, the process of automatically filling in the regional data divided into time periods in this stage according to the extended load curve axis includes several key steps. First, the data of the extended load curve axis is mapped with the regional data divided into time periods in the current stage. This step determines the matching path between them by comparing the time series shapes of the two data sets to ensure that the load data of the extended load curve axis is accurately mapped to the current time period area.

[0116] For example, assuming that the extended load curve axis during a certain historical shutdown period shows a morning load peak and an afternoon load trough, and assuming that the current time-divided regional data includes daily hourly load records, the peak of the extended load curve axis may appear between 6 and 8 o'clock, while the trough may appear between 14 and 16 o'clock. By mapping these peak and trough data of the extended load curve axis with the current time-divided data, the load level of the historical peak can be applied to the corresponding time period in the current data, forming a load forecasting model consistent with the actual load demand.

[0117] By comparing the sequence shapes during the mapping process, the changing trends and periodic characteristics of the load data can be identified. For example, if the load change trends in certain seasons or holidays are found in the historical data to be consistent with the data in the current stage, adjustments can be made based on these trends. For example, assuming that the historical data has a more obvious load peak in winter, then in the mapping process, the load forecast value for the winter period needs to be increased accordingly to reflect this seasonal change.

[0118] Finally, after the mapping is completed, these adjusted data are combined with the load data of the current stage to generate accurate load forecasting results. This method improves the accuracy of load forecasting by comparing and mapping historical data with current data, and can better cope with changes in load demand in different time periods.

[0119] In a preferred embodiment, a time-divided power grid load forecasting system includes an analysis module that mines sub-region data after time-divided and analyzes the sub-region data;

[0120] The prediction module uses an integrated learning algorithm to divide the decomposed sub-region data into components and uses a time series prediction model to combine the decomposition results and the component results;

[0121] The filling module collects the grid load factor during the historical shutdown period, automatically fills in the regional data divided by time period in this stage according to the extended load curve axis, and predicts the grid load based on the automatically filled data.

[0122] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0123] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0124] In summary, the present invention effectively solves the shortcomings of traditional prediction methods in dealing with multi-dimensional factors and data missing problems through multi-model integration algorithms and automatic filling technology. It uses the improved Spearman correlation coefficient and statistical bootstrap method to accurately estimate the load temperature and improve the prediction accuracy and robustness based on the time series prediction model. In addition, for special circumstances such as shutdown periods, the automatic filling of the load curve axis is extended to ensure the continuity and accuracy of time-divided load forecasting, thereby improving the reliability and efficiency of power system operation management.

[0125] Example 2

[0126] Reference Figure 1 and Figure 2 , which is the second embodiment of the present invention, and this embodiment provides a method for predicting power grid load by time period division. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0127] During the experiment, the grid load forecasting technical solution was comprehensively tested to verify its accuracy and reliability. The experiment selected a data set including multiple different grid areas and time periods, covering the actual grid load data for the whole year of 2023. The data set includes hourly load records and related meteorological data, such as temperature and humidity.

[0128] The data for 2023 were preprocessed and the data for the whole year were divided into different time periods, such as 6:00-8:00 in the morning, 12:00-14:00 at noon, 18:00-20:00 in the evening, and 22:00-24:00 at night. For example, the load data of a certain area on the morning of July 15, 2023 was 500MW, 800MW at noon, 1000MW in the evening, and 700MW at night. Then, the improved Spearman correlation coefficient was used to calculate the load temperature in each time period, and the temperature sensitivity analysis results were obtained.

[0129] During the data analysis stage, the statistical bootstrap method was used to estimate the strong load temperature and weak load temperature. Specifically, multiple sampling samples were generated by resampling the data 1000 times, and the strong load temperature and weak load temperature of the samples were statistically analyzed. The results showed that the mean of the strong load temperature was 23℃, and the standard deviation was 2℃, while the mean of the weak load temperature was 18℃, and the standard deviation was 1.5℃. Based on these statistics, the Fourier decomposition algorithm was applied to decompose the load data into daily period component, periodic component, low-frequency component, high-frequency component and constant term, and the SVM, Light GBM and Ridge algorithms were used for ensemble learning of the low-frequency components, with weights of 3:6:1 respectively.

[0130] The Prophet algorithm is used to predict the time series of periodic components, high-frequency components and constant terms. The prediction results show that the power grid load on January 15, 2024 will reach 550MW in the morning, 820MW at noon, 1050MW in the evening, and 730MW at night. In order to verify these prediction results, the data of the extended load curve axis is mapped with the regional data divided by the current time period. By comparing the sequence shape of the load data during the mapping process, the load data in the morning and evening match the peaks and troughs in the historical data well.

[0131] The experimental results show that these technical means can accurately predict future load conditions, provide reliable data support for power grid load management, optimize power grid resource allocation, and improve the accuracy and practicality of prediction. The power grid load data prediction is shown in Table 2 below:

[0132] Table 2 Power grid load data forecast table

[0133] date Time Load data Forecast Load High load temperature Weak load air temperature July 15, 2023 Morning 6:00-8:00 500 550 23 18 July 15, 2023 12:00-14:00 noon 800 820 23 18 July 15, 2023 Evening 18:00-20:00 1000 1050 23 18 July 15, 2023 Night 22:00-24:00 700 730 23 18 Prediction for January 15, 2024 Morning 6:00-8:00 0 550 23 18 Prediction for January 15, 2024 12:00-14:00 noon 0 820 23 18 Prediction for January 15, 2024 Evening 18:00-20:00 0 1050 23 18 Prediction for January 15, 2024 Night 22:00-24:00 0 730 23 18

[0134] The table shows the key data of the power grid load forecasting experiment, including actual load data and forecast results. Specifically, the table lists the actual power grid loads in different time periods on July 15, 2023, 500MW in the morning, 800MW at noon, 1000MW in the evening, and 700MW at night, as well as the load forecast values ​​for the corresponding time periods on January 15, 2024 obtained by the forecasting model, 550MW in the morning, 820MW at noon, 1050MW in the evening, and 730MW at night. In addition, the table also includes statistical data on strong load temperature and weak load temperature. The mean of the strong load temperature is 23°C, the standard deviation is 2°C, and the mean of the weak load temperature is 18°C, and the standard deviation is 1.5°C. These data show the actual application effect and accuracy of load forecasting during the experiment. The comparison between the present invention and the prior art is shown in Table 3 below:

[0135] Table 3 Comparison between the present invention and the prior art

[0136]

[0137] The present invention has obvious advantages in load forecasting technology. Compared with the existing technology, the present invention uses the improved Spearman correlation coefficient and the bootstrap method to dynamically estimate the impact of temperature on load, and combines the Fourier decomposition algorithm with the integrated learning model to perform data decomposition and prediction. Compared with the traditional method, the present invention significantly improves the prediction accuracy through double time period division and dynamic load factor correction. At the same time, the technology of automatically filling and correcting load data further improves the accuracy and practicality of the prediction.

[0138] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A time-divided power grid load forecasting method, characterized in that: include, Mining the sub-regional data after time division, analyzing the sub-regional data, and using the combined load forecasting algorithm to decompose the analyzed sub-regional data; The decomposed sub-region data are divided into components using an integrated learning algorithm, and the decomposition results and component results are combined using a time series prediction model; Collect the grid load factor during historical shutdown periods, predict the extended load curve axis of the grid load factor, automatically fill in the regional data divided into time periods in this stage according to the extended load curve axis, and predict the grid load based on the automatically filled data.

2. The time-divided power grid load forecasting method according to claim 1, characterized in that: The mining of the sub-region data divided into time periods includes calculating the overall grid load factor of the sub-region data according to the improved Spearman correlation coefficient, the overall grid load factor includes calculating the load temperature of the grid load, and optimizing the overall grid load factor based on the load temperature; The specific calculation formula of the load air temperature is: Among them, xi represents the candidate characteristic value of the overall power grid load factor, represents the mean value of the candidate feature value, yi represents the load value divided in this time period, represents the average value of load value, r represents the load temperature; When the absolute value |r|→1, the attribute correlation of the load temperature is enhanced; When the absolute value |r|→0, the attribute correlation of the load temperature is weakened.

3. The time-divided power grid load forecasting method according to claim 2, characterized in that: The analysis of the sub-region data includes estimating the high-load air temperature and the low-load air temperature according to the statistical bootstrap method, wherein the estimation includes repeatedly sampling the overall power grid load factor to generate multiple sampling samples, and obtaining the statistics of the load air temperature according to the generated multiple sampling samples; Decomposing the analyzed sub-region data includes dynamically programming the statistics of the load temperature and defining the Ferecher distance, wherein the definition of the Ferecher distance includes defining it according to the difference of the load temperature counted on the day; The specific calculation formula for the repeated sampling is: Where n represents the sample size, Indicates the high load temperature, Indicates the weak load temperature, represents the strong load temperature of the i-th observation in the b-th resampled sample, represents the weak load temperature of the i-th observation in the b-th resampled sample, represents the standard deviation of the strong load temperature in the bth resampled sample, represents the standard deviation of the weak load air temperature in the bth resampled sample.

4. The time-divided power grid load forecasting method according to claim 3 is characterized in that: Decomposing the analyzed sub-region data includes decomposing the statistics based on the Fourier decomposition algorithm, and the decomposition results of the statistics include a daily cycle component, a cycle component, a low-frequency component, a high-frequency component and a constant term. The specific calculation formula of the decomposition is: y(t)=g(t)+s(t)+h(t)+ε t Among them, y(t) represents the statistical result data after decomposition, g(t) represents the trend component, s(t) represents the periodic component, h(t) represents the date impact component, ε t represents the residual term, t represents the time point; The low-frequency component is used as the temperature sensitive component, and the temperature sensitive component includes the SVM algorithm, the Light GBM algorithm and the Ridge algorithm, which are weighted in a ratio of 3:6:1, and the integrated learning model is input with the forecast temperature of the day to obtain the prediction result; The periodic component, high-frequency component and constant term are added together as the time series component, and the time series load forecasting is completed based on the Prophet algorithm; The specific calculation formula of the SVM algorithm is: Among them, f(x) represents the regression prediction output value of the SVM algorithm, α i represents the Lagrange multiplier, K(x i , x represents the kernel function, x i represents the candidate characteristic value of the overall power grid load factor, x represents the input characteristic vector to be predicted, and b represents the resampled sample; The specific calculation formula of the Light GBM algorithm is: Among them, F(x) represents the regression prediction output value of the Light GBM algorithm, γ m represents the weight coefficient of the mth weak learner, T m (x) represents the output of the mth decision tree, and M represents the total number of weak learners; The specific calculation formula of the Ridge algorithm is: Among them, J(θ) represents the regression prediction output value of the Ridge algorithm, yi represents the load value divided in this time period, θ represents the regression coefficient vector, and θTx i represents the predicted value of the regression model, λ represents the regularization parameter, represents the square of the regression coefficient; The specific calculation formula of the Prophet algorithm is: Among them, φ represents the regression prediction output value of the Prophet algorithm, and y(t) represents the statistical result data after decomposition.

5. The time-divided power grid load forecasting method according to claim 4 is characterized in that: The merging of the decomposition results and the component results includes connecting the decomposed sub-region data into a curve for smoothing, horizontally stretching the curve according to the stage divided at this time, dividing it into time periods for the second time, and constructing the load coefficient of the stage divided at this time according to the change ratio of the new and old coefficients every day; If the load factor of the first time period division is greater than the load factor of the second time period division, the future predicted load factor will be corrected according to the average error rate of the previous day's time period division. Otherwise, all predicted load factors are aggregated to obtain the total predicted load before and after the stage divided at this time.

6. The time-divided power grid load forecasting method according to claim 5, characterized in that: The extended load curve axis for predicting the load factor of the power grid includes preprocessing the transverse stretching and selecting the accuracy of the time period division according to the stretching ratio; If there are load peaks or troughs during the historical downtime period, the stretch ratio is determined by using a smoothing method of local expansion or contraction; If there is no load change feature, the default stretch ratio is used; if there is a trough during the historical downtime period, the time series during the historical downtime period is analyzed to identify the long-term trend and seasonal changes of the time series.

7. The time-divided power grid load forecasting method according to claim 6, characterized in that: The automatically filling in the regional data divided by time periods in the current stage according to the extended load curve axis includes mapping the extended load curve axis with the regional data divided by time periods in the current stage; The data of the extended load curve axis is mapped to the regional data divided into the current time periods. By comparing the sequence shapes of the two during the mapping process, the matching path is found to complete the power grid load forecast.

8. A time-divided power grid load forecasting system, based on the time-divided power grid load forecasting method according to any one of claims 1 to 7, characterized in that: include, An analysis module, which mines the sub-region data divided into time periods and analyzes the sub-region data; The prediction module uses an integrated learning algorithm to divide the decomposed sub-region data into components and uses a time series prediction model to combine the decomposition results and the component results; The filling module collects the grid load factor during the historical shutdown period, automatically fills in the regional data divided by time period in this stage according to the extended load curve axis, and predicts the grid load based on the automatically filled data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting power grid load by time period division as claimed in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting power grid load by time period division as claimed in any one of claims 1 to 7 are implemented.

Images