Coordinated load prediction method and device, electronic equipment and storage medium
By employing multi-dimensional data fusion and dual-dimensional correction methods, the problems of low accuracy in load forecasting and poor model interpretability were solved, enabling accurate forecasting of complex and variable load scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING TUNING TECHNOLOGY CO LTD
- Filing Date
- 2026-03-04
- Publication Date
- 2026-06-12
Smart Images
Figure CN122196363A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power system dispatching, and in particular to a method, apparatus, electronic device and storage medium for predicting load under unified dispatch. Background Technology
[0002] Currently, the mainstream technical solutions for load forecasting include: 1) traditional time series forecasting techniques represented by models such as ARIMA (Autoregressive Integral Moving Average) and exponential smoothing; 2) machine learning-driven load forecasting techniques represented by neural networks and random forests; 3) load forecasting techniques based on matching similar days; and 4) forecasting techniques based on load characteristic classification. These forecasting solutions still suffer from low forecasting accuracy, poor model interpretability, and difficulty in adapting to complex and ever-changing actual load scenarios.
[0003] Therefore, it is necessary to provide a unified load forecasting scheme with high prediction accuracy, high model interpretability, and adaptability to complex and ever-changing actual load scenarios to solve the above-mentioned problems of existing technologies. Summary of the Invention
[0004] In view of this, embodiments of this application provide a method, apparatus, electronic device and storage medium for predicting load under unified dispatch, in order to solve the problems of low prediction accuracy, poor model interpretability and difficulty in adapting to complex and ever-changing actual load scenarios in existing load prediction schemes.
[0005] A first aspect of this application provides a method for predicting loads under unified dispatch, including: The historical load data, meteorological numerical model data and provincial characteristic parameters are preprocessed to obtain the preprocessed historical load data, preprocessed meteorological numerical model data and preprocessed provincial characteristic parameters. Determine the target forecast date and its corresponding target date attribute label. Based on the target date attribute label, select a set of similar dates from the preprocessed historical load data. The set of similar dates includes at least one similar date, and the similar dates have the same basic load characteristics as the target forecast date. Based on the historical load curve corresponding to each similar date, generate the baseline value curve of the unified dispatch load corresponding to the target prediction date; Based on the preprocessed meteorological numerical model data, determine the meteorological correction value corresponding to the target forecast date; Based on the preprocessed provincial characteristic parameters, determine the provincial characteristic correction value corresponding to the target prediction date; Based on meteorological correction values and provincial characteristic correction values, the baseline curve of the unified dispatch load is corrected in two dimensions, and the final unified dispatch load forecast curve corresponding to the target forecast date is output.
[0006] A second aspect of this application provides a load forecasting device, comprising: The preprocessing module is configured to preprocess historical load data, meteorological numerical model data and provincial characteristic parameters to obtain preprocessed historical load data, preprocessed meteorological numerical model data and preprocessed provincial characteristic parameters. The filtering module is configured to determine the target forecast date and its corresponding target date attribute label, and to filter out a set of similar dates from the preprocessed historical load data based on the target date attribute label. The set of similar dates includes at least one similar date, and the similar dates have the same basic load characteristics as the target forecast date. The generation module is configured to generate a baseline curve of the unified dispatch load corresponding to the target prediction date based on the historical load curve corresponding to each similar date. The first determination module is configured to determine the meteorological correction value corresponding to the target forecast date based on the preprocessed meteorological numerical model data. The second determining module is configured to determine the province characteristic correction value corresponding to the target prediction date based on the preprocessed province characteristic parameters. The output module is configured to perform a two-dimensional correction on the baseline curve of the unified dispatch load based on meteorological correction values and provincial characteristic correction values, and output the final unified dispatch load forecast curve corresponding to the target forecast date.
[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.
[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.
[0009] Compared with the prior art, the beneficial effects of this application embodiment include at least the following: by multi-dimensional data fusion (fusion of historical unified dispatch load data, meteorological numerical model data and provincial characteristic parameters), dynamic similar day screening (pre-classification based on target date attribute labels, and accurate selection of a set of similar dates with high reference value from the pre-processed historical unified dispatch load data), and dual-dimensional correction (dual-dimensional accurate correction of the unified dispatch load base curve based on meteorological correction values and provincial characteristic correction values), the final unified dispatch load prediction curve is obtained, which effectively solves the problems of low prediction accuracy, poor model interpretability, and difficulty in adapting to complex and ever-changing actual load scenarios in the prior art. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a flowchart illustrating a load forecasting method for unified dispatch provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a load forecasting device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0013] The following will describe in detail, with reference to the accompanying drawings, a method and apparatus for predicting load according to a unified dispatch system, based on an embodiment of this application.
[0014] Load forecasting for the entire province is a core foundation for power system dispatching and operation, and resource optimization. Currently, the mainstream technical solutions for load forecasting include: (1) traditional time series forecasting techniques represented by ARIMA (Autoregressive Integral Moving Average) and exponential smoothing models; (2) machine learning-driven load forecasting techniques represented by neural networks and random forests; (3) load forecasting techniques based on similar day matching; and (4) forecasting techniques based on load characteristic classification. These forecasting schemes still have the following shortcomings: First, the characterization of load drivers is insufficient and imprecise. None of the above schemes (1) to (4) fully and deeply explored the relationship between load and influencing factors. For example, scheme (1) completely ignores core driving factors such as meteorological and date attributes; scheme (2) only includes a few meteorological elements and does not consider the coupling effect of meteorological factors and the sensitivity differences in different time periods; schemes (3) and (4) only apply meteorological data superficially and do not use multidimensional meteorological data for fine correction.
[0015] Second, the ability to adapt to dynamic changes is insufficient. None of the above schemes (1) to (4) fully consider the dynamic variables in the operation of the power grid, especially the changes in the installed capacity of power sources (mainly new energy sources). When the installed capacity of power sources in the region is significantly adjusted, the load carrying capacity and characteristics of the power grid will change accordingly. However, the existing technology still uses historical data or simple mean correction, which leads to the distortion of the prediction basis.
[0016] Third, the differentiation and modeling of different load scenarios are inadequate. The above-mentioned scheme (1) does not differentiate between the load characteristics of weekdays and holidays, and uses a unified model to process the load data of weekdays, weekends and statutory holidays. However, the load characteristics of reduced industrial production activities and increased residential electricity consumption during holidays are significantly different from those of weekdays, which further exacerbates the prediction error. Scheme (2) does not perform differentiated modeling for the differences in meteorological sensitivity during daytime / nighttime periods, and cannot accurately adapt to the differences in load patterns under different scenarios. The classification dimensions (season, time period) of schemes (3) and (4) are relatively simple, and the correction methods are crude, which cannot accurately adapt to the differences in load patterns under different scenarios.
[0017] Fourth, the ability to cope with extreme and sudden load scenarios is weak. The above-mentioned schemes (1) to (4) are all unable to capture load changes caused by extreme weather (sustained high temperature, cold wave, etc.) and cannot cope with the situation where the load peak exceeds the historical extreme value. In such scenarios, the prediction deviation increases significantly.
[0018] In summary, existing load forecasting schemes still suffer from low forecasting accuracy, poor model interpretability, and difficulty in adapting to complex and ever-changing actual load scenarios.
[0019] In view of this, the embodiments of this application provide a method for predicting the unified dispatch load. Through multi-dimensional data fusion (integrating historical unified dispatch load data, meteorological numerical model data, and provincial characteristic parameters), dynamic similar day screening (pre-classifying based on target date attribute labels to accurately select a set of similar dates with high reference value from the preprocessed historical unified dispatch load data), and dual-dimensional correction (performing dual-dimensional accurate correction of the unified dispatch load base curve based on meteorological correction values and provincial characteristic correction values), the final unified dispatch load prediction curve is obtained. This method effectively solves the problems of low prediction accuracy, poor model interpretability, and difficulty in adapting to complex and ever-changing actual load scenarios in existing technologies.
[0020] Figure 1 This is a flowchart illustrating a load forecasting method provided in an embodiment of this application. Figure 1 The unified load forecasting method can be executed by the server. For example... Figure 1 As shown, the load forecasting method includes the following steps: Step S101: Preprocess the historical load data, meteorological numerical model data, and provincial characteristic parameters to obtain preprocessed historical load data, preprocessed meteorological numerical model data, and preprocessed provincial characteristic parameters.
[0021] Historical load data refers to the time-series historical dataset formed by the power grid dispatching agency continuously collecting all grid-connected power loads within its unified dispatching jurisdiction at a specific time resolution, after cleaning, verification, and labeling with spatiotemporal and related attributes.
[0022] For example, historical load data can be monthly historical load data with a resolution of 96 points from a provincial-level dispatch center. This data meets the following characteristics: 1) Dispatch scope: covers all grid-connected users (industrial, residential, commercial, and public utility) in the province, excluding self-generated and self-consumed loads of self-owned power plants within the province; 2) Time resolution: 96 points (15 minutes / point), including 96 sampling points from 00:00, 00:15 to 23:45 on a single day, and a total of 2976 data points for the entire month (31 days); 3) Data content: each sampling point corresponds to the total load value of the dispatch center (unit: MW), and is labeled with data quality tags (normal / abnormal / missing) and date attribute tags (e.g., October 1st is labeled "National Day holiday"); 4) Related information: includes statistics on the average load of working days, average load of holidays, and peak load for the month, and is also associated with daily average temperature, humidity, and other meteorological data.
[0023] Meteorological numerical model data refers to a mathematical model (i.e., meteorological numerical model) built based on physical laws such as atmospheric dynamics and thermodynamics. It uses real-time observed meteorological data as the initial field and boundary conditions, and simulates and predicts a gridded, spatiotemporally continuous, multi-dimensional meteorological element dataset by solving a set of differential equations of atmospheric motion using computer numerical methods.
[0024] Meteorological numerical model data, including temperature, humidity, wind speed, solar radiation intensity, and precipitation probability.
[0025] Matching the spatiotemporal resolution of meteorological numerical model data with load data, that is, the temporal granularity and spatial coverage of meteorological numerical model data must correspond completely with the temporal and spatial dimensions of load data, in order to ensure the accuracy of characterizing the impact of meteorological factors on load. This is a key prerequisite for improving the accuracy of load forecasting.
[0026] The matching of the spatiotemporal resolution of meteorological numerical model data with load data can be understood from two dimensions: ① The temporal resolution of the meteorological numerical model data must match the load data, meaning the sampling interval of the meteorological numerical model data must be consistent with the sampling interval of the load data. For example, if the load data has a resolution of 96 points (15 minutes / point), the meteorological numerical model data used for analysis must also be hourly data at the 15-minute level (such as temperature and solar radiation intensity every 15 minutes) to accurately capture short-term linkage patterns such as "a sudden increase in solar radiation within a certain 15 minutes → an increase in air conditioning load". ② The spatial resolution of the meteorological numerical model data must match the load data, meaning the grid coverage of the meteorological numerical model data must be consistent with the statistical range of the load. For example, if the load is the total load of the provincial unified dispatch, meteorological data using a 3-5 kilometer grid covering the entire province is sufficient, without needing to be refined to the urban street level.
[0027] Provincial characteristic parameters refer to a set of quantitative indicators used to characterize the core attributes of a province's power system operation, load characteristics, energy structure, and industrial layout. These parameters have regional stability and scenario adaptability, and are key basic data for building provincial unified dispatch load forecasting models, formulating power grid dispatching strategies, and planning power generation capacity.
[0028] Provincial characteristic parameters include industrial structure weights, typical electricity consumption curves, and holiday response patterns.
[0029] Among them, the industrial structure weight refers to the proportion of the electricity load of different industries (or industrial categories) to the total electricity load of the region. It is a core quantitative indicator for measuring the degree of influence of industrial structure on electricity load, and directly determines the load change pattern, peak and valley characteristics and growth trend.
[0030] Typical electricity consumption curves refer to the standardized load change trajectory obtained by statistically analyzing the electricity load of a certain region, user group, or industry type over a time dimension (minute, hour, day, week, month). They intuitively reflect the peak-valley characteristics, fluctuation patterns, and time distribution characteristics of electricity load, and are the core basis for analyzing load characteristics and formulating dispatching strategies.
[0031] Holiday response mode refers to a standardized operating procedure developed by power grid dispatching agencies for non-working day scenarios such as statutory holidays and adjusted workdays, adapting to changes in load characteristics in load forecasting, supply and demand balancing, demand-side response, and emergency handling. Its core function is to address sudden changes in load structure caused by industrial shutdowns and altered travel or consumption habits during holidays, ensuring the safe and stable operation of the power grid.
[0032] The core characteristics of provincial characteristic parameters include: A) Regional specificity: They are strongly correlated with the province's industrial structure, climate conditions, population distribution, etc., and the parameter values differ significantly between different provinces; B) Relative stability: They do not change drastically in the short term and are usually updated on an annual or quarterly basis, reflecting the long-term operating rules of the province's power system; C) Multi-dimensional coverage: They cover multiple dimensions such as load characteristics, energy structure, industrial electricity consumption, and meteorological sensitivity coefficients, supporting full-scenario analysis of the power system.
[0033] As an example, historical load data can be collected through the internal system of the power grid dispatching agency, meteorological numerical model data can be collected through professional meteorological agencies and the power grid's self-developed meteorological platform, and provincial characteristic parameters of each province can be obtained through multi-department collaborative collection and internal statistical analysis of the power grid.
[0034] Preprocessing is performed on historical load data, meteorological numerical model data, and provincial characteristic parameters. Specifically, this includes labeling the differential characteristics of these three types of data using a combination of manual and machine labeling. For example, manual labeling is used to annotate abnormal data and special events (such as "peak summer load control") in historical load data, while machine labeling is used to automatically annotate routine attributes (such as date type: weekday, weekend, statutory holiday, adjusted workday; load peak and valley labels: morning peak, evening peak, valley) in historical load data. Another example is the manual labeling of extreme weather events (such as "rainstorm warning") and meteorological element coupled scenarios (such as "high temperature and high humidity") in meteorological numerical model data; simultaneously, machine labeling is used to automatically annotate the threshold labels of image elements in meteorological numerical model data (such as: temperature ≥35℃ (high temperature), precipitation probability ≥80% (heavy rain)). For example, manual labeling is used to mark the industrial electricity consumption characteristics (e.g., high-energy-consuming industrial concentrated areas) and holiday response patterns (e.g., low load during the Spring Festival) in the provincial characteristic parameters; at the same time, machine labeling is used to automatically mark the threshold of characteristic parameters in the provincial characteristic parameters (e.g., industrial electricity consumption ratio ≥ 60% (major industrial province)).
[0035] Furthermore, historical load data and meteorological numerical model data are aligned according to timestamps, and missing values are filled using interpolation. Specifically, the timestamps of historical load data can be used as a baseline, and the timestamps of meteorological numerical model data can be aligned with the baseline. Missing values are then filled using interpolation. For example, if only historical load data is available but no meteorological numerical model data is available, or vice versa, the missing values can be fitted using valid data before and after the missing values through mathematical methods, ensuring the continuity and rationality of the data.
[0036] By integrating multi-dimensional data such as load, meteorology, and provincial characteristics, and performing preprocessing such as data alignment, interpolation completion, and outlier removal, data quality and utilization can be improved.
[0037] Step S102: Determine the target forecast date and its corresponding target date attribute label. Based on the target date attribute label, select a set of similar dates from the preprocessed historical load data. The set of similar dates includes at least one similar date, and the basic load characteristics of the similar dates are consistent with those of the target forecast date.
[0038] Step S103: Based on the historical load curve corresponding to each similar date, generate the load baseline curve corresponding to the target prediction date.
[0039] Step S104: Based on the preprocessed meteorological numerical model data, determine the meteorological correction value corresponding to the target forecast date.
[0040] Step S105: Based on the preprocessed provincial characteristic parameters, determine the provincial characteristic correction value corresponding to the target prediction date.
[0041] Step S106: Based on meteorological correction values and provincial characteristic correction values, perform dual-dimensional correction on the baseline curve of the unified dispatch load, and output the final unified dispatch load forecast curve corresponding to the target forecast date.
[0042] The technical solution provided in this application, through multi-dimensional data fusion (fusion of historical unified dispatch load data, meteorological numerical model data, and provincial characteristic parameters), dynamic similar day screening (pre-classification based on target date attribute labels, accurately selecting a set of similar dates with high reference value from the pre-processed historical unified dispatch load data), and dual-dimensional correction (dual-dimensional accurate correction of the unified dispatch load base curve based on meteorological correction values and provincial characteristic correction values), obtains the final unified dispatch load prediction curve. This effectively solves the problems of low prediction accuracy, poor model interpretability, and difficulty in adapting to complex and ever-changing actual load scenarios in existing technologies.
[0043] In some embodiments, a set of similar dates is selected from the preprocessed historical load data based on the target date attribute label, including: From the preprocessed historical load data, a set of candidate dates corresponding to the target date attribute label is selected. The set of candidate dates includes multiple candidate dates. Calculate the load characteristic similarity between each candidate date and the target predicted date, and based on the load characteristic similarity, filter out the set of similar dates from the candidate date set.
[0044] As an example, if the target predicted date If the corresponding target date attribute label is "working day", then all historical dates with the date attribute label "working day" (i.e., candidate dates) are selected from the preprocessed historical load data. This ensures that the candidate dates match the target prediction date. The basic load characteristics are consistent.
[0045] The higher the similarity of load characteristics, the stronger the candidate date. With target forecast date The more similar the load characteristics between them, the better; conversely, the lower the similarity of load characteristics, the better the candidate date. With target forecast date The more dissimilar the load characteristics are between them.
[0046] In some embodiments, calculating the load characteristic similarity between each candidate date and the target predicted date includes: For each candidate date, calculate the meteorological similarity and time decay function value between the candidate date and the target prediction date; Determine the first weighting coefficient corresponding to the similarity of meteorological characteristics, and the second weighting coefficient corresponding to the time decay function value; The load characteristic similarity between the candidate date and the target prediction date is determined based on meteorological characteristic similarity, the first weighting coefficient, the time decay function value, and the second weighting coefficient.
[0047] As an example, candidate dates are calculated according to formulas (1) and (2). With target forecast date The similarity of meteorological characteristics between them.
[0048] (1); In equation (1), Indicates candidate date With target forecast date The similarity of meteorological characteristics between them Indicates candidate date With target forecast date The standardized Euclidean distance between the corresponding meteorological elements (temperature, humidity, wind speed, solar radiation intensity, and precipitation probability). The smaller the standardized Euclidean distance, the higher the similarity of the meteorological characteristics of the two; conversely, the larger the standardized Euclidean distance, the lower the similarity of the meteorological characteristics of the two.
[0049] (2); In equation (2), n represents the number of meteorological elements. For example, if the meteorological elements include temperature, humidity, wind speed, solar radiation intensity, and precipitation probability, then n = 5. Indicates candidate date The standardized value of the k-th meteorological element; Indicates the target prediction date The standardized value of the kth meteorological element.
[0050] Calculate the candidate date according to formula (3) With target forecast date The time decay function value between.
[0051] (3); In equation (3), Indicates candidate date With target forecast date The time decay function value between; Indicates the time interval between the candidate date and the target predicted date; for example, candidate date. October 1, 2022, is the target forecast date. If it is September 28, 2022, then... It is 3 days; This indicates the maximum time span, which can be flexibly set according to the actual situation, such as 30 days.
[0052] In the scenario of matching similar days in load forecasting, the core function of the time decay function is to give higher similarity weights to historical dates that are closer to the target forecast date, and lower similarity weights to historical dates that are further away, thereby reflecting the higher reference value of recent load patterns.
[0053] Calculate the candidate date according to formula (4) With target forecast date The similarity of load characteristics between them.
[0054] (4); In equation (4), Indicates candidate date With target forecast date Similarity of load characteristics between them; This represents the first weighting coefficient. Denotes the second weighting coefficient, where It is obtained through back-optimization of historical prediction errors.
[0055] Each candidate date is calculated using the formulas (1) to (4) above. With target forecast date The similarity of load characteristics between dates is calculated, and then the dates are sorted from high to low according to the similarity of load characteristics. Multiple candidate dates with high similarity of load characteristics are selected as similar dates to obtain a set of similar dates.
[0056] The technical solution provided in this application adopts a dynamic similar day filtering rule with date attribute prior classification, which avoids confusion of different types of date load characteristics; at the same time, it combines a time decay mechanism to highlight the reference value of recent similar days and ensure the accuracy of similar days.
[0057] In some embodiments, based on the historical load curve corresponding to each similar date, a baseline curve of the centrally dispatched load corresponding to the target forecast date is generated, including: For each similar date, determine its corresponding weighting coefficient based on the similarity of the load characteristics corresponding to the similar dates; The load curves corresponding to each similar date are weighted and calculated to generate the baseline load curve corresponding to the target forecast date.
[0058] As an example, each similar date is calculated according to formula (5). The corresponding weighting coefficients.
[0059] (5); In equation (5), Indicates similar dates The corresponding weighting coefficients; Indicates similar dates Corresponding load characteristic similarity; This represents the sum of the load characteristic similarities of all similar dates in the set of similar dates.
[0060] Construct the load baseline function corresponding to the target forecast date, as shown in formula (6).
[0061] (6); In equation (6), This represents the load baseline function; m represents the total number of similar dates in the set of similar dates; Similar dates The corresponding weighting coefficients; Indicates similar dates Corresponding historical load function (and similar dates) (corresponding to the historical load curve).
[0062] Next, based on the load baseline function corresponding to the target forecast date, the corresponding load baseline curve is generated.
[0063] By using the above method, the load characteristics of all highly similar dates can be integrated, laying the foundation for subsequent two-dimensional accurate correction.
[0064] In some embodiments, the preprocessed meteorological numerical model data includes at least basic thermal elements, dynamic elements, radiation elements and precipitation elements, and the spatiotemporal resolution of the preprocessed meteorological numerical model data matches the load data. Based on preprocessed meteorological numerical model data, meteorological correction values corresponding to the target forecast date are determined, including: The preprocessed meteorological numerical model data is input into a pre-trained load-meteorological sensitivity model, which outputs meteorological correction values corresponding to the target forecast date; the load-meteorological sensitivity model is a fitting model based on a backpropagation neural network.
[0065] As an example, the training steps for the load-meteorological sensitivity model are as follows: First, collect historical meteorological numerical model data (including temperature, humidity, wind speed, sunshine intensity, precipitation probability, etc.); then, divide the historical meteorological numerical model data into training set, validation set, and test set according to a preset ratio; use the training set to train the initial backpropagation neural network model (BP model) until the preset convergence condition is reached, then terminate the training to obtain the BP training model; use the validation set and test set to validate and test the BP training model. If the validation results and test results both meet the preset requirements (such as the model accuracy reaching a preset threshold), then the trained load-meteorological sensitivity model is obtained.
[0066] The mathematical expression for the load-meteorological sensitivity model is shown in equation (7): (7); In equation (7), Indicates the weather correction value; The model is a fitting model based on backpropagation (BP) neural network; T represents temperature, Q represents humidity, W represents wind speed, P represents precipitation probability, and S represents solar radiation intensity.
[0067] In practical applications, the target meteorological numerical model data corresponding to the target forecast date is first extracted from the preprocessed meteorological numerical model data. Then, the target meteorological numerical model data is input into the trained load-meteorological sensitivity model, and the meteorological correction value corresponding to the target forecast date is output.
[0068] The above scheme uses preprocessed multidimensional meteorological numerical model data (temperature, humidity, wind speed, solar radiation intensity, precipitation probability, etc.) as model input. The meteorological data has achieved precise spatiotemporal resolution matching with the load data. Combined with the strong fitting capability of the backpropagation neural network for nonlinear relationships, it can effectively characterize the load response patterns under the coupled effects of multiple meteorological elements, overcoming the limitations of traditional techniques that rely on single meteorological factors or manual empirical coefficient corrections. For complex meteorological scenarios such as high temperature and high humidity, low temperature and strong winds, and extreme weather scenarios, it can reduce load forecasting errors and solve the technical pain point of insufficient forecasting accuracy under complex meteorological conditions using traditional methods.
[0069] Furthermore, by automatically learning the correlation between meteorological elements and load correction values in historical data through backpropagation neural networks, the model eliminates the need for manually setting empirical coefficients and frequent parameter adjustments based on actual operating conditions, significantly reducing labor costs and human error. Simultaneously, the model supports offline training and online prediction modes, seamlessly integrating with existing power grid load forecasting systems. The output meteorological correction values can be directly used to optimize basic load forecasting results, enhancing the automation and intelligence of the load forecasting process.
[0070] In some embodiments, determining the province characteristic correction value corresponding to the target prediction date based on the preprocessed province characteristic parameters includes: Extract the target load correction pattern corresponding to the target forecast date from the preprocessed provincial characteristic parameters; Determine the provincial characteristic influence intensity coefficient, which reflects the degree of influence of provincial characteristic parameters on the overall dispatch load; Based on the target load correction model and the provincial characteristic influence intensity coefficient, determine the provincial characteristic correction value corresponding to the target forecast date.
[0071] The value range of the provincial characteristic influence intensity coefficient is [0,1]. This coefficient reflects the degree of influence of provincial characteristics on the overall load and needs to be verified and determined in conjunction with historical data. For example, the characteristic intensity coefficient for the Spring Festival holiday is set to 0.95, and that for a regular weekend is set to 0.7.
[0072] The larger the coefficient, the greater the influence of the provincial characteristic parameter on the overall load; conversely, the smaller the coefficient, the smaller the influence of the provincial characteristic parameter on the overall load.
[0073] As an example, first, determine the target forecast date. Corresponding date attribute labels are used to clarify the type of the target prediction date, such as weekdays (peak summer season) and public holidays (Chinese New Year). Then, the target prediction date is extracted from the preprocessed provincial characteristic parameters. Matching target load adjustment pattern. Specifically, according to the target forecast date. The corresponding date attribute tags retrieve the corresponding target load correction mode, such as the midday industrial load reduction percentage (%), the evening residential load peak increase (MW), and the commercial load flat-period correction ratio. For example, the target forecast date... The corresponding date attribute label is the Spring Festival holiday, and the target load correction mode is a decrease of a% in midday industrial load and an increase of b MW in peak evening residential load. Next, the provincial characteristic influence intensity coefficient corresponding to this date attribute label is obtained from the parameter library. Finally, the provincial characteristic correction value corresponding to the target prediction date is calculated according to formula (8).
[0074] (8); In equation (8), Indicates the province characteristic correction value; This represents the intensity coefficient of the influence of provincial characteristics; Indicates the date of the target forecast. Matching target load correction mode.
[0075] This invention introduces a configurable provincial characteristic parameter library and couples the provincial characteristic intensity coefficient with the load correction mode, enabling personalized dynamic adjustment of the load curve. It can quickly adapt to the needs of provinces with different industrial structures and electricity consumption characteristics without rebuilding the model. It can effectively cope with the load forecasting needs of complex scenarios such as holidays, extreme weather, and seasonal changes, and solves the technical problem of large differences in forecasting accuracy of traditional general models in different provinces.
[0076] In some embodiments, the above method further includes: Based on meteorological correction values and provincial characteristic correction values, the baseline curve of the unified dispatch load is corrected in two dimensions to obtain the preliminary forecast curve of the unified dispatch load. The preliminary load forecast curve is validated for rationality, and the final load forecast curve corresponding to the target forecast date is output according to the preset time resolution.
[0077] As an example, based on the meteorological correction value, provincial characteristic correction value, and unified dispatch load base value function obtained in the above embodiments, a preliminary unified dispatch load prediction function is constructed, and its mathematical expression is shown in equation (9).
[0078] (9); In equation (9), This represents the preliminary load forecasting function (corresponding to the preliminary load forecasting curve). Represents the load baseline function for unified regulation; This indicates a weather correction value.
[0079] Next, the rationality of the preliminary load forecast curve corresponding to the preliminary load forecast function will be verified to eliminate loads that exceed the historical load fluctuation range. The system calculates outliers and outputs the final load forecast curve corresponding to the target forecast date, based on the preset time resolution (e.g., 24 or 96 points) for the actual province.
[0080] In summary, the beneficial effects of the technical solutions provided in this application include: (1) This application innovatively introduces the date distance decay factor into the similar date evaluation system and constructs a dynamic similar date screening algorithm, so that the closer the similar date is to the target prediction date, the higher the reference weight is obtained, which accurately matches the time-series dependence characteristics of the actual load, effectively improves the rationality and accuracy of similar date screening, and solves the technical problems of traditional similar date screening not considering time-series weight and unbalanced reference value.
[0081] (2) This application constructs a load-meteorological sensitivity model based on backpropagation (BP) neural network. Through this model, the coupling correlation analysis of multiple meteorological elements such as temperature, humidity, wind speed, and solar radiation intensity is realized and the load forecast is accurately corrected. This breaks through the technical limitations of the existing technology, which simplifies the meteorological correction method and cannot capture the synergistic effect of multiple meteorological elements on the load, and improves the forecast reliability in complex meteorological scenarios.
[0082] (3) This application establishes a configurable provincial characteristic parameter library, integrates core parameters such as industrial structure, electricity consumption characteristics, and climate adaptability of different provinces, and realizes personalized dynamic adjustment of load curves through the correlation and coupling of provincial characteristic intensity coefficient and load correction mode, thereby achieving differentiated customization of load forecast, effectively improving the adaptability of forecast model between provinces, and solving the problems of poor adaptability and uneven forecast accuracy of traditional general models in different provinces.
[0083] (4) This application adopts a modular design approach to construct a three-level load forecasting framework with clear hierarchy (a three-level forecasting framework of "base value superposition - meteorological correction - characteristic correction"). The load base value provides the overall load change trend within the forecast period, the meteorological correction module captures the impact of short-term meteorological fluctuations on the load, and the characteristic correction module embeds the long-term electricity consumption patterns of the province. The three are organically combined to form a complete and closed-loop load forecasting link, taking into account both forecasting accuracy and model maintainability. Furthermore, the physical meaning of each module in the forecasting process, such as similar day selection, meteorological correction, and characteristic correction, is clear, making it easy for operation and maintenance personnel to understand and adjust.
[0084] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.
[0085] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.
[0086] Figure 2 This is a schematic diagram of a load forecasting device provided in an embodiment of this application. Figure 2 As shown, the load forecasting device 200 includes: The preprocessing module 201 is configured to preprocess historical load data, meteorological numerical model data and provincial characteristic parameters to obtain preprocessed historical load data, preprocessed meteorological numerical model data and preprocessed provincial characteristic parameters. The filtering module 202 is configured to determine the target forecast date and its corresponding target date attribute label, and to filter out a set of similar dates from the preprocessed historical load data based on the target date attribute label. The set of similar dates includes at least one similar date, and the similar date has the same basic load characteristics as the target forecast date. The generation module 203 is configured to generate a load baseline curve corresponding to the target prediction date based on the historical load curve corresponding to each similar date. The first determining module 204 is configured to determine the meteorological correction value corresponding to the target forecast date based on the preprocessed meteorological numerical model data. The second determining module 205 is configured to determine the province characteristic correction value corresponding to the target prediction date based on the preprocessed province characteristic parameters. Output module 206 is configured to perform a two-dimensional correction on the baseline curve of the unified dispatch load based on meteorological correction values and provincial characteristic correction values, and output the final unified dispatch load forecast curve corresponding to the target forecast date.
[0087] In some embodiments, the filtering module 202 described above includes: The first filtering unit is configured to filter out a set of candidate dates corresponding to the target date attribute label from the preprocessed historical load data; the set of candidate dates includes multiple candidate dates. The second filtering unit is configured to calculate the load characteristic similarity between each candidate date and the target predicted date, and to filter out a set of similar dates from the candidate date set based on the load characteristic similarity.
[0088] In some embodiments, the second filtering unit described above includes: The calculation component is configured to calculate the meteorological similarity between the candidate date and the target prediction date and the time decay function value for each candidate date. The first determining component is configured to determine a first weighting coefficient corresponding to the similarity of meteorological characteristics, and a second weighting coefficient corresponding to the value of the time decay function; The second determining component is configured to determine the load characteristic similarity between the candidate date and the target prediction date based on meteorological characteristic similarity, a first weighting coefficient, a time decay function value, and a second weighting coefficient.
[0089] In some embodiments, the generation module 203 described above may be specifically configured as follows: For each similar date, determine its corresponding weighting coefficient based on the similarity of the load characteristics corresponding to the similar dates; The load curves corresponding to each similar date are weighted and calculated to generate the baseline load curve corresponding to the target forecast date.
[0090] In some embodiments, the preprocessed meteorological numerical model data includes at least basic thermal elements, dynamic elements, radiation elements, and precipitation elements, and the spatiotemporal resolution of the preprocessed meteorological numerical model data matches the load data. The aforementioned first determining module 204 can be specifically configured as follows: The preprocessed meteorological numerical model data is input into a pre-trained load-meteorological sensitivity model, which outputs meteorological correction values corresponding to the target forecast date; the load-meteorological sensitivity model is a fitting model based on a backpropagation neural network.
[0091] In some embodiments, the second determining module 205 described above may be specifically configured as follows: Extract the target load correction pattern corresponding to the target forecast date from the preprocessed provincial characteristic parameters; Determine the provincial characteristic influence intensity coefficient, which reflects the degree of influence of provincial characteristic parameters on the overall dispatch load; Based on the target load correction model and the provincial characteristic influence intensity coefficient, determine the provincial characteristic correction value corresponding to the target forecast date.
[0092] In some embodiments, the output module 206 described above may be specifically configured as follows: Based on meteorological correction values and provincial characteristic correction values, the baseline curve of the unified dispatch load is corrected in two dimensions to obtain the preliminary forecast curve of the unified dispatch load. The preliminary load forecast curve is validated for rationality, and the final load forecast curve corresponding to the target forecast date is output according to the preset time resolution.
[0093] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0094] Figure 3 This is a schematic diagram of the electronic device 300 provided in an embodiment of this application. For example... Figure 3 As shown, the electronic device 300 of this embodiment includes a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various device embodiments described above.
[0095] Electronic device 300 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 300 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 300 and does not constitute a limitation on electronic device 300. It may include more or fewer parts than shown, or different parts.
[0096] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
[0097] The memory 302 can be an internal storage unit of the electronic device 300, such as a hard disk or RAM of the electronic device 300. The memory 302 can also be an external storage device of the electronic device 300, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device 300. The memory 302 can also include both internal and external storage units of the electronic device 300. The memory 302 is used to store computer programs and other programs and data required by the electronic device.
[0098] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device 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 embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0099] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.
[0100] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A method for predicting load under unified dispatch, characterized in that, include: The historical load data, meteorological numerical model data and provincial characteristic parameters are preprocessed to obtain the preprocessed historical load data, preprocessed meteorological numerical model data and preprocessed provincial characteristic parameters. Determine the target prediction date and its corresponding target date attribute label, and based on the target date attribute label, filter out a set of similar dates from the preprocessed historical load data. The set of similar dates includes at least one similar date, and the similar date has the same basic load characteristics as the target prediction date. Based on the historical load curve corresponding to each of the similar dates, a baseline load curve corresponding to the target prediction date is generated; Based on the preprocessed meteorological numerical model data, a meteorological correction value corresponding to the target forecast date is determined; Based on the preprocessed provincial characteristic parameters, determine the provincial characteristic correction value corresponding to the target prediction date; Based on the meteorological correction value and the provincial characteristic correction value, the baseline curve of the unified dispatch load is corrected in two dimensions, and the final unified dispatch load forecast curve corresponding to the target forecast date is output.
2. The method according to claim 1, characterized in that, Based on the target date attribute label, a set of similar dates is selected from the preprocessed historical load data, including: From the preprocessed historical load data, a set of candidate dates corresponding to the target date attribute label is selected, and the set of candidate dates includes multiple candidate dates. Calculate the load characteristic similarity between each candidate date and the target predicted date, and based on the load characteristic similarity, filter out a set of similar dates from the candidate date set.
3. The method according to claim 2, characterized in that, Calculating the load characteristic similarity between each candidate date and the target predicted date includes: For each candidate date, calculate the meteorological characteristic similarity and time decay function value between the candidate date and the target predicted date; Determine a first weighting coefficient corresponding to the similarity of the meteorological characteristics, and a second weighting coefficient corresponding to the time decay function value; The load characteristic similarity between the candidate date and the target prediction date is determined based on the meteorological characteristic similarity, the first weighting coefficient, the time decay function value, and the second weighting coefficient.
4. The method according to claim 1, characterized in that, Based on the historical load curve corresponding to each of the similar dates, a baseline load curve corresponding to the target forecast date is generated, including: For each of the similar dates, a weighting coefficient is determined based on the load characteristic similarity corresponding to the similar date. Based on the historical load curves and weighting coefficients corresponding to each of the similar dates, a weighted calculation is performed to generate the baseline load curve corresponding to the target prediction date.
5. The method according to claim 1, characterized in that, The preprocessed meteorological numerical model data includes at least basic thermal elements, dynamic elements, radiation elements, and precipitation elements, and the spatiotemporal resolution of the preprocessed meteorological numerical model data matches the load data. Based on the preprocessed meteorological numerical model data, determine the meteorological correction value corresponding to the target forecast date, including: The preprocessed meteorological numerical model data is input into a pre-trained load-meteorological sensitivity model, which outputs a meteorological correction value corresponding to the target forecast date; wherein the load-meteorological sensitivity model is a fitting model based on a backpropagation neural network.
6. The method according to claim 1, characterized in that, Based on the preprocessed provincial characteristic parameters, determine the provincial characteristic correction value corresponding to the target prediction date, including: Extract the target load correction pattern corresponding to the target forecast date from the preprocessed provincial characteristic parameters; Determine the provincial characteristic influence intensity coefficient, which reflects the degree of influence of provincial characteristic parameters on the central dispatch load; Based on the target load correction mode and the provincial characteristic influence intensity coefficient, determine the provincial characteristic correction value corresponding to the target forecast date.
7. The method according to claim 1, characterized in that, Based on the meteorological correction value and the provincial characteristic correction value, the baseline curve of the unified dispatch load is corrected in two dimensions, and the final unified dispatch load forecast curve corresponding to the target forecast date is output, including: Based on the meteorological correction value and the provincial characteristic correction value, the baseline curve of the unified dispatch load is corrected in two dimensions to obtain the preliminary unified dispatch load forecast curve. The preliminary load forecast curve is validated for rationality, and the final load forecast curve corresponding to the target forecast date is output according to the preset time resolution.
8. A load forecasting device for centralized dispatch, characterized in that, include: The preprocessing module is configured to preprocess historical load data, meteorological numerical model data and provincial characteristic parameters to obtain preprocessed historical load data, preprocessed meteorological numerical model data and preprocessed provincial characteristic parameters. The filtering module is configured to determine the target prediction date and its corresponding target date attribute label, and to filter out a set of similar dates from the preprocessed historical load data based on the target date attribute label. The set of similar dates includes at least one similar date, and the similar date has the same basic load characteristics as the target prediction date. The generation module is configured to generate a load baseline curve corresponding to the target prediction date based on the historical load curve corresponding to each of the similar dates. The first determining module is configured to determine a meteorological correction value corresponding to the target forecast date based on the preprocessed meteorological numerical model data. The second determining module is configured to determine a province characteristic correction value corresponding to the target prediction date based on the preprocessed province characteristic parameters. The output module is configured to perform a two-dimensional correction on the baseline curve of the unified dispatch load based on the meteorological correction value and the provincial characteristic correction value, and output the final unified dispatch load forecast curve corresponding to the target forecast date.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.