Power sales prediction method and device based on multi-feature decomposition, equipment and medium

Abstract

The application discloses a kind of based on multi-feature decomposition's electricity sales volume prediction method, device, equipment and medium, the method includes: obtaining the electricity data and exogenous influence data of target area, and the electricity data and exogenous influence data are preprocessed, obtain the standardized target electricity sequence;Based on target electricity sequence and exogenous influence data, exogenous feature is constructed, and target electricity sequence is decomposed in multiple ways, and multiple scale sequence component is obtained;According to the feature of each multi-scale sequence component, the component prediction is carried out to the corresponding prediction model, and each component prediction result is obtained, and each component prediction result is synthesized according to decomposition rule, and initial electricity sales volume prediction result is obtained;Based on rolling prediction mechanism, initial electricity sales volume prediction result is evaluated and selected, and final electricity sales volume prediction result is obtained.The technical scheme can improve the prediction accuracy of electricity sales volume, enhance the business adaptability and stability of prediction result.

Description

Technical Field

[0001] This invention relates to the field of electricity sales forecasting technology, and in particular to a method, apparatus, equipment and medium for forecasting electricity sales based on multi-feature decomposition. Background Technology

[0002] Electricity sales forecasting is a core foundation for power grid planning, dispatching, operation, and business decisions. Its forecasting accuracy directly affects the efficiency of power grid resource allocation and operational benefits.

[0003] Currently, most existing electricity sales forecasting methods use a single model to predict the overall original electricity sales sequence, failing to fully consider the multi-scale characteristics of the electricity sales sequence itself, such as trends, seasonality, and random disturbances. Furthermore, they do not effectively quantify the impact of exogenous factors such as the Spring Festival and abnormal weather on electricity sales, resulting in poor adaptability of the forecasting models to complex electricity consumption scenarios. At the same time, existing methods lack rolling evaluation and selection mechanisms tailored to power grid operations, and the forecast results often fail to meet actual business performance requirements, exhibiting problems such as low forecast accuracy, weak business applicability, and insufficient stability, thus failing to provide reliable support for precise power grid dispatching and scientific decision-making. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for predicting electricity sales based on multi-feature decomposition, in order to improve the accuracy of electricity sales prediction and enhance the business adaptability and stability of the prediction results.

[0005] According to one aspect of the present invention, a method for predicting electricity sales based on multi-feature decomposition is provided, comprising: Acquire the power data and exogenous influence data of the target area, and preprocess the power data and exogenous influence data to obtain a standardized target power sequence; Based on the target power sequence and the exogenous influence data, exogenous features are constructed, and the target power sequence is decomposed in multiple ways to obtain multi-scale sequence components. Based on the feature matching of each multi-scale sequence component, the corresponding prediction model is used to predict the components, and the prediction results of each component are obtained. Then, the prediction results of each component are synthesized according to the decomposition rules to obtain the initial electricity sales prediction result. The initial electricity sales forecast results are evaluated and selected based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

[0006] According to another aspect of the present invention, a power sales forecasting device based on multi-feature decomposition is provided, comprising: The data acquisition and preprocessing module is used to acquire the power data and exogenous influence data of the target area, and preprocess the power data and exogenous influence data to obtain a standardized target power sequence. The feature construction and decomposition module is used to construct exogenous features based on the target electricity sequence and the exogenous influence data, and to perform multi-mode feature decomposition on the target electricity sequence to obtain multi-scale sequence components; The component prediction and synthesis module is used to perform component prediction based on the feature matching of each multi-scale sequence component with the corresponding prediction model, obtain the prediction results of each component, and synthesize the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result. The evaluation and selection module is used to evaluate and select the initial electricity sales forecast results based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

[0007] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to execute the electricity sales forecasting method based on multi-feature decomposition according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute and implement the electricity sales prediction method based on multi-feature decomposition of any embodiment of the present invention.

[0009] The technical solution of this invention acquires electricity data and exogenous influence data of a target area, preprocesses the electricity data and exogenous influence data to obtain a standardized target electricity sequence; constructs exogenous features based on the target electricity sequence and exogenous influence data, and performs multi-mode feature decomposition on the target electricity sequence to obtain multi-scale sequence components; performs component prediction by matching the features of each multi-scale sequence component with the corresponding prediction model, obtains the prediction results of each component, and synthesizes the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result; evaluates and selects the initial electricity sales prediction result based on a rolling prediction mechanism to obtain the final electricity sales prediction result. This technical solution solves the technical problems of insufficient multi-scale time series feature mining, insufficient quantification of exogenous influencing factors, poor adaptability of prediction models, and difficulty in meeting the rolling assessment requirements of power grid business in existing electricity sales prediction methods. It improves the accuracy and stability of electricity sales prediction, enhances the engineering applicability of the solution, and provides reliable data support for precise planning and efficient operation of the power grid.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a method for predicting electricity sales based on multi-feature decomposition, provided in an embodiment of the present invention; Figure 2 A schematic diagram of a power sales forecasting device based on multi-feature decomposition provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an electronic device for implementing the electricity sales prediction method based on multi-feature decomposition in this embodiment of the invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Figure 1This is a flowchart illustrating a multi-feature decomposition-based electricity sales forecasting method provided in an embodiment of the present invention. This embodiment is applicable to situations requiring accurate electricity sales forecasting. The method can be executed by a multi-feature decomposition-based electricity sales forecasting device, which can be implemented in hardware and / or software and can be configured in a computer device. Figure 1 As shown, the method specifically includes the following steps: S110. Obtain the power data and exogenous influence data of the target area, and preprocess the power data and exogenous influence data to obtain a standardized target power sequence.

[0016] The target area can be the region where electricity sales forecasting is carried out, the electricity data can be statistical data representing the electricity consumption and sales situation in the region, the exogenous influence data can be relevant data that have an external impact on electricity sales, and the standardized target electricity sequence can be an electricity sequence that meets the requirements of subsequent modeling after unified processing.

[0017] Specifically, power data and exogenous influence data of the target area can be obtained, and preprocessing operations can be performed on the above data to remove anomalies and unify the format, forming a standardized target power sequence.

[0018] S120. Based on the target power sequence and the exogenous influence data, construct exogenous features and perform multi-mode feature decomposition on the target power sequence to obtain multi-scale sequence components.

[0019] Among them, exogenous features can be quantitative features constructed based on external influencing factors, multi-mode feature decomposition can be a processing method that splits the sequence using multiple different principles, and multi-scale sequence components can be data components with different change patterns and time scales obtained after decomposition.

[0020] Specifically, exogenous features can be constructed based on the target power sequence and exogenous influence data, and multi-mode feature decomposition can be performed on the target power sequence to obtain multi-scale sequence components.

[0021] S130. Based on the feature matching of each multi-scale sequence component, the corresponding prediction model is used to predict the components, and the prediction results of each component are obtained. The prediction results of each component are then combined according to the decomposition rules to obtain the initial electricity sales prediction result.

[0022] The prediction model can be a mathematical or machine learning model used for time series data prediction, the component prediction can be the process of predicting each decomposed sequence separately, the decomposition rule can be the reconstruction rule corresponding to the decomposition method, and the initial electricity sales prediction result can be the preliminary prediction value.

[0023] Specifically, based on the variation characteristics of each multi-scale sequence component, a suitable prediction model can be matched to complete the component prediction, and the prediction results of each component can be obtained. Then, the prediction results of each component can be synthesized according to the rules corresponding to the decomposition method to obtain the initial electricity sales prediction result.

[0024] S140. The initial electricity sales forecast results are evaluated and selected based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

[0025] Among them, the rolling forecasting mechanism can be to dynamically update data and carry out forecasting according to the power grid business cycle, the evaluation and selection can be the process of verifying the error of the forecasting results and determining the optimal solution, and the final electricity sales forecasting result can be the electricity sales forecasting value that meets business constraints and has the best accuracy.

[0026] Specifically, the initial electricity sales forecast results can be evaluated and selected based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

[0027] In some possible implementations, the preprocessing of the electricity data and exogenous influence data to obtain a standardized target electricity sequence includes: converting the cumulative electricity sales and consumption data into corresponding data for the current month; performing missing value imputation processing based on the electricity consumption characteristics of the corresponding missing month when the electricity data is missing; and unifying all processed electricity data to the unit of ten thousand kilowatt-hours to obtain the standardized target electricity sequence.

[0028] Among them, the cumulative electricity sales and consumption data can be electricity data obtained by accumulating statistics over time, while the corresponding data for the current month can be data that only represents the electricity consumption and sales situation for a single month; missing value imputation can be an operation to fill in the missing parts of the data, and electricity consumption business characteristics can be business attributes related to the monthly electricity consumption pattern.

[0029] Specifically, when preprocessing electricity data and exogenous influence data, cumulative electricity sales and consumption data can be converted to monthly data to ensure data period consistency. In cases of missing electricity data, imputation can be performed based on the electricity consumption characteristics of the missing month. Finally, all processed electricity data is standardized to the unit of ten thousand kilowatt-hours, resulting in a standardized target electricity sequence.

[0030] In some possible implementations, when the electricity data is missing, the missing value imputation process is performed based on the electricity consumption characteristics of the corresponding missing month, including: when the data for the beginning month is missing, data imputation is performed based on the percentage of working days in the corresponding month; when the data for the end month is missing, regression imputation or equivalent imputation is performed based on the historical electricity consumption patterns of the same period.

[0031] Among them, the beginning month can be the first month of each year, the percentage of working days can be the percentage of working days in the corresponding month, the end month can be the last month of each year, the historical electricity consumption pattern can be the electricity consumption change pattern of the same period in history, regression interpolation can be a way to supplement data based on regression models, and equivalent interpolation can be a way to supplement data based on historical equivalent relationships.

[0032] In this embodiment of the invention, when performing missing value imputation, if data for the beginning month is missing, data imputation is performed based on the percentage of working days in the corresponding month to match the electricity consumption fluctuation characteristics at the beginning of the year. If data for the end month is missing, regression imputation or equivalent imputation can be used based on historical electricity consumption patterns during the same period to ensure that the missing data matches actual business patterns. This differentiated imputation method can improve data restoration accuracy and reduce sources of prediction error.

[0033] In some possible implementations, constructing exogenous features based on the target electricity consumption sequence and the exogenous influence data includes: using electricity consumption data from the electricity consumption data as the core explanatory feature; constructing a continuously quantified Spring Festival intensity feature based on the actual number of days the Spring Festival covers in the corresponding month; constructing meteorological disturbance features based on the number of days with abnormally high and low temperatures in the region; and integrating the core explanatory feature, the Spring Festival intensity feature, and the meteorological disturbance feature to obtain the exogenous feature. The exogenous features, in conjunction with the target electricity consumption sequence, are used to perform multi-mode feature decomposition and component prediction on the target electricity consumption sequence. Core explanatory features can be key features with strong explanatory power for electricity sales; the Spring Festival intensity feature can be a quantitative feature characterizing the degree of impact of the Spring Festival on the electricity consumption of the current month; the actual coverage days can be the actual number of days of the Spring Festival holiday within the current month; the meteorological disturbance feature can be a quantitative feature characterizing the impact of abnormal temperatures on electricity consumption; abnormal high temperatures and abnormal low temperatures can be temperature conditions that exceed the normal range.

[0034] Understandably, when constructing exogenous features, electricity consumption data from the electricity data can be used as the core explanatory feature. A continuous, quantified Spring Festival intensity feature can be constructed based on the actual number of days the Spring Festival covers the corresponding month. Meteorological disturbance features can be constructed based on the number of days with abnormally high and low temperatures in the region. These three types of features are then integrated to form a complete exogenous feature. This exogenous feature is used in conjunction with the target electricity series for subsequent multi-mode feature decomposition and component prediction. This feature construction method comprehensively incorporates key external influencing factors, improving the model's predictive ability for special months.

[0035] For example, the changes in production / lifestyle patterns caused by major holidays such as the Spring Festival are quantitatively coded. Since the Spring Festival holiday often spans multiple months, using only a 0 / 1 variable indicating whether the month includes the Spring Festival can easily miss cross-month information and introduce discontinuities. Therefore, a holiday effect variable is constructed based on monthly coverage ratios. Let the number of days covered by the Spring Festival holiday in month t be , and the holiday length be uniformly 7 days. Then, the monthly Spring Festival intensity variable Ht = dt / 7, dt ∈ {0, 1, ..., 7} is defined. This variable can simultaneously reflect whether the Spring Festival occurs and its coverage degree within the current month.

[0036] In some possible implementations, the multi-mode feature decomposition of the target energy sequence to obtain multi-scale sequence components includes: decomposing the target energy sequence using X13 seasonal adjustment decomposition to obtain trend cyclic components, seasonal components, and irregular components; decomposing the target energy sequence using discrete wavelet decomposition to obtain low-frequency trend components and high-frequency detail components; adaptively decomposing the target energy sequence using empirical mode decomposition to obtain intrinsic mode function components and residual trend components; and integrating the above decomposition results to obtain the multi-scale sequence components.

[0037] Among them, X13 seasonally adjusted decomposition is a time series decomposition method based on the principle of seasonal adjustment. The trend cycle component is the component that represents the long-term trend and cycle change. The seasonal component is the component that represents the monthly seasonal change. The irregular component is the component that represents the random disturbance. Discrete wavelet decomposition is a multi-scale decomposition method based on wavelet transform. The low-frequency trend component is the component that represents the overall slow change. The high-frequency detail component is the component that represents the short-term rapid fluctuation. Empirical mode decomposition is an adaptive time series decomposition method. The intrinsic mode function component is the intrinsic mode component obtained by decomposition. The residual trend component is the remaining overall trend term after decomposition.

[0038] It should be noted that when performing multi-mode eigenvalue decomposition, the target energy quantity sequence can be decomposed into trend cyclic components, seasonal components, and irregular components through X13 seasonal adjustment decomposition; into low-frequency trend components and high-frequency detail components through discrete wavelet decomposition; and into intrinsic mode function components and residual trend components through empirical mode decomposition and adaptive decomposition. Integrating these multiple decomposition results forms a complete multi-scale sequence component. The complementary use of multiple decomposition methods enhances the adaptability to sequences with different fluctuation characteristics.

[0039] In some possible implementations, the step of performing component prediction based on the characteristic matching of each multi-scale sequence component with a corresponding prediction model includes: for linear stationary multi-scale sequence components, using an autoregressive distributed lag model for component prediction; for multi-scale sequence components with seasonal periodicity, using a seasonal time series model with exogenous variables for component prediction; for nonlinear high-frequency multi-scale sequence components, using a long short-term memory network model for component prediction; through the above targeted prediction processing, the component prediction results corresponding to each multi-scale sequence component are obtained, and the component prediction results are synthesized according to decomposition rules to obtain the initial electricity sales prediction result.

[0040] Among them, linear stationary multi-scale sequence components can be components with stable and linear changes; autoregressive distributed lag models can be linear prediction models based on autoregressive and lagged variables; multi-scale sequence components with seasonal periodicity can be components with fixed seasonal periodic changes; seasonal time series models with exogenous variables can be seasonal time series prediction models incorporating external features; nonlinear high-frequency multi-scale sequence components can be components with drastic changes and nonlinear relationships; and long short-term memory network models can be deep learning models suitable for time-dependent relationships.

[0041] Specifically, when predicting components, for linear stationary multi-scale sequence components, an autoregressive distributed lag model is used for prediction. The model input is historical stationary sequence data and related features, and the output is the predicted value of the component. For multi-scale sequence components with seasonal periodicity, a seasonal time series model with exogenous variables is used for prediction. The model input is historical seasonal sequences and exogenous features, and the output is the predicted value of the component. For nonlinear high-frequency multi-scale sequence components, a long short-term memory network model is used for prediction. The model input is historical high-frequency time series data, and the output is the predicted value of the component. The prediction results of each component are obtained through the above targeted predictions, and then synthesized according to the decomposition rules to obtain the initial electricity sales prediction result. The categorized modeling fully adapts to the characteristics of each component, significantly improving the overall prediction accuracy.

[0042] In some possible implementations, the evaluation and selection of the initial electricity sales forecast results based on the rolling forecast mechanism to obtain the final electricity sales forecast results includes: performing rolling forecasts on an annual basis, updating the training data with the latest monthly actual electricity data in each round of rolling forecasts; calculating the annual electricity sales forecast error corresponding to each round of rolling forecasts; selecting forecast results whose errors in all rounds meet preset constraints; and selecting the model with the smallest average error from the compliant models to complete the selection and obtain the final electricity sales forecast results.

[0043] Among them, the year can be the complete natural year cycle used for forecasting, rolling forecasting can be the method of updating the latest data and re-conducting forecasts, training data can be historical data used for model training, annual electricity sales forecasting error can be the deviation between the annual forecast value and the actual value, preset constraints can be the error limit conditions stipulated by the power grid business, compliant model can be a forecasting model that meets all error constraints, and average error can be the average of multiple forecasting errors.

[0044] In this embodiment of the invention, when performing evaluation and selection based on the rolling forecasting mechanism, rolling forecasts are performed on an annual cycle. Each round of rolling forecasting updates the training data with the latest monthly actual electricity consumption data. The forecasting error for the corresponding annual electricity sales is calculated round by round, and forecasting results where the error in all rounds meets preset constraints are selected. The model with the smallest average error among the compliant models is chosen to complete the final selection, and the final electricity sales forecast result is output. This mechanism aligns with the actual business processes of the power grid, ensuring that the forecasting results meet actual assessment requirements.

[0045] The technical solution of this invention acquires electricity data and exogenous influence data of a target area, preprocesses the electricity data and exogenous influence data to obtain a standardized target electricity sequence; constructs exogenous features based on the target electricity sequence and exogenous influence data, and performs multi-mode feature decomposition on the target electricity sequence to obtain multi-scale sequence components; performs component prediction by matching the features of each multi-scale sequence component with the corresponding prediction model, obtains the prediction results of each component, and synthesizes the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result; evaluates and selects the initial electricity sales prediction result based on a rolling prediction mechanism to obtain the final electricity sales prediction result. This technical solution solves the technical problems of insufficient multi-scale time series feature mining, insufficient quantification of exogenous influencing factors, poor adaptability of prediction models, and difficulty in meeting the rolling assessment requirements of power grid business in existing electricity sales prediction methods. It improves the accuracy and stability of electricity sales prediction, enhances the engineering applicability of the solution, and provides reliable data support for precise planning and efficient operation of the power grid.

[0046] In an optional embodiment, feature decomposition, component prediction, result synthesis, rolling prediction and evaluation, model selection and output can be performed through the following process: Feature decomposition module (multi-decomposition method library) The feature decomposition module is used to perform one or more time-series decomposition processes on the electricity sales sequence and the optional electricity consumption sequence, outputting multiple interpretable subsequence components, specifically including the following decomposition methods: (1) X13 Seasonal Adjustment Decomposition Before performing X13 seasonal adjustment, the original sequence is preprocessed with RegARIMA to eliminate the interference of special factors such as trading days and holidays, as well as abnormal fluctuations, on the seasonal decomposition.

[0047] Specifically, the X13 program first establishes a regression-ARIMA extended model: Here, exogenous regression variables such as holiday effects and trading day effects are represented, and the residual sequence is the result of regression adjustment. A seasonal ARIMA(p,d,q)(P,D,Q)s model is then fitted to the residuals to characterize the autocorrelation and seasonal structure of the sequence. The model orders p,q,P,Q are usually automatically selected based on the AIC / BIC information criterion, while the difference orders d,D are determined through stationarity tests.

[0048] After preprocessing and outlier correction, X13 further employs the X-11 iterative moving average filtering method, or a seasonal adjustment method based on the ARIMA model (derived from the SEATS seasonal adjustment procedure), to decompose the series into a trend term T, a seasonal term S, and an irregularity term I, thus obtaining the seasonally adjusted result, supporting both additive and multiplicative forms. In engineering applications, the trend and period can be combined into a TC component for subsequent analysis and forecasting.

[0049] (2) DWT Discrete Wavelet Decomposition The filter bank is used to obtain low-frequency trends and high-frequency details / noise components, and supports different decomposition levels. In this project, two decomposition levels were selected from the perspective of comparison with seasonal adjustment. At the same time, based on the data length, three decomposition levels were selected. Sym3 and CoIF5 wavelet bases were selected respectively to obtain good reconstruction results. Sym3 is suitable for lightweight and high-efficiency scenarios, while CoIF5 is suitable for high-precision and high-smoothness scenarios.

[0050] (3) EMD empirical mode decomposition The target electrical quantity sequence is adaptively decomposed to obtain several IMF components and residual trend terms; each IMF satisfies the extreme point / zero crossing point constraint and the upper and lower envelope mean is zero constraint, and it is iterated until SD≤0.2–0.3 or the residual is monotonic; the high-frequency IMFs can be further merged by similarity, and sequences with similar trends can be merged to reduce the number of components and facilitate modeling and synthesis.

[0051] The components obtained from the above decomposition methods are integrated to form multi-scale sequence components, which are used for subsequent component prediction.

[0052] Component prediction module (multi-prediction model library) The component prediction module is used to train prediction models for each component obtained from the decomposition and then execute predictions. It uses data from 2006 to 2022 to train the models and then uses data from 2023 to test the model performance. It supports at least the following models: (1) ADL Autoregressive Distributed Lag Model Using electricity sales as the dependent variable and electricity consumption as the exogenous variable, autoregression and distributed lag terms are introduced to model the model from the perspectives of model significance and the time lag relationship of actual data.

[0053] (2) Seasonal time series model with exogenous variables in SARIMAX To enhance predictive performance, exogenous variables are introduced into seasonal ARIMA. This can be expressed as: in, For exogenous regression terms, It can include variables such as the intensity of the Spring Festival and abnormal temperatures; the remaining part is characterized by the seasonal ARIMA structure, thereby improving the model's ability to explain and predict special shocks while retaining seasonality and autocorrelation characteristics.

[0054] (3) LSTM Long Short-Term Memory Network Model This tool is used to capture long dependencies in time series data. Configurable parameters include the number of network layers, units, optimizer, batch size, and rolling window cross-validation. Optimizer: adam (learning rate 0.001); Batch size: 32; Training epochs: 50 (loss function: MSE); Random seed: 42 for both NumPy and TensorFlow (to ensure reproducibility); Cross-validation: A rolling window strategy is used, with the training set adding one data point at a time, splitting the data into 12 windows in chronological order.

[0055] For different types of multi-scale sequence components, a matching prediction model is used to perform component prediction, and the prediction results for each component are obtained.

[0056] 7 Result Synthesis Module The results synthesis module is used to reconstruct the predicted results of each component into the predicted electricity sales value according to the decomposition rules. The specific synthesis methods include: (1) For the X13 additive model: the predicted electricity sales volume is obtained by adding the prediction results of TC, S and I components; For the X13 multiplicative model: the components are synthesized according to the multiplicative relationship.

[0057] (2) For DWT: synthesize the low-frequency component prediction and the high-frequency component prediction of each layer according to the reconstruction relationship.

[0058] (3) For EMD: The residual trend is added to the combined IMF component prediction results.

[0059] Using the above method, the prediction results of each component are combined into the initial electricity sales prediction result.

[0060] Rolling Forecasting and Evaluation Module (Business Process Oriented) The rolling forecasting and evaluation module performs forecasting and evaluation according to the "intra-year rolling update" business method of power grid companies: 12 rolling forecasts are required each year from January to December; the first forecast uses December of the previous year as the training cutoff point to predict January to December of the current year; each subsequent forecast adds the latest actual month to the training set, only updates the forecasts for subsequent months, and keeps the forecasts for the already predicted months unchanged, and then summarizes them to obtain the annual forecast.

[0061] Meanwhile, based on the power grid company's requirement that "the absolute relative error of the annual forecast should not exceed 2%", an evaluation standard is proposed: in the 12 rolling forecasts, the error corresponding to each annual forecast must meet the above evaluation standard. In engineering practice, this can be implemented by determining and counting the error of each rolling forecast.

[0062] Model selection and output module The model selection and output module is used to compare and automatically select combinations of "decomposition methods × prediction models," and output the final recommended model and prediction results. The device can be pre-loaded with a library of "multiple decomposition methods × multiple prediction models." For example, X13, DWT, and EMD can be combined with ADL, SARIMAX, and LSTM to form multiple candidate models, so as to compare the differences between decomposition methods and model types simultaneously.

[0063] The output may include: monthly forecasts, annual forecasts, error assessment reports, explanations of trend / seasonal / high-frequency disturbance components, and reasons for model recommendations. For example, EMD or DWT can be selected based on regional volatility.

[0064] Figure 2 This is a schematic diagram of a power sales forecasting device based on multi-feature decomposition, provided as an embodiment of the present invention. Figure 2 As shown, the device includes: The data acquisition and preprocessing module 210 is used to acquire the power data and exogenous influence data of the target area, and preprocess the power data and exogenous influence data to obtain a standardized target power sequence. The feature construction and decomposition module 220 is used to construct exogenous features based on the target power sequence and the exogenous influence data, and to perform multi-mode feature decomposition on the target power sequence to obtain multi-scale sequence components; The component prediction and synthesis module 230 is used to perform component prediction based on the feature matching of each multi-scale sequence component with the corresponding prediction model, obtain the prediction results of each component, and synthesize the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result. The evaluation and selection module 240 is used to evaluate and select the initial electricity sales forecast results based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

[0065] The technical solution of this invention acquires electricity data and exogenous influence data of a target area, preprocesses the electricity data and exogenous influence data to obtain a standardized target electricity sequence; constructs exogenous features based on the target electricity sequence and exogenous influence data, and performs multi-mode feature decomposition on the target electricity sequence to obtain multi-scale sequence components; performs component prediction by matching the features of each multi-scale sequence component with the corresponding prediction model, obtains the prediction results of each component, and synthesizes the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result; evaluates and selects the initial electricity sales prediction result based on a rolling prediction mechanism to obtain the final electricity sales prediction result. This technical solution solves the technical problems of insufficient multi-scale time series feature mining, insufficient quantification of exogenous influencing factors, poor adaptability of prediction models, and difficulty in meeting the rolling assessment requirements of power grid business in existing electricity sales prediction methods. It improves the accuracy and stability of electricity sales prediction, enhances the engineering applicability of the solution, and provides reliable data support for precise planning and efficient operation of the power grid.

[0066] In some possible implementations, the data acquisition and preprocessing module 210 includes: The caliber conversion submodule is used to convert the cumulative electricity sales and consumption data into the corresponding data for the current month. The missing value interpolation submodule is used to perform missing value interpolation processing based on the electricity consumption characteristics of the corresponding missing month when the electricity data is missing. The dimension unification submodule is used to unify all processed electricity data to the dimension of ten thousand kilowatt-hours to obtain the standardized target electricity sequence.

[0067] In some possible implementations, the missing value imputation submodule includes: The beginning-of-year interpolation unit is used to perform data interpolation based on the percentage of working days in the corresponding month when data for the beginning of the year is missing. The year-end interpolation unit is used to perform regression interpolation or equivalent interpolation based on historical electricity consumption patterns during the same period when data for the last month of the year is missing.

[0068] In some possible implementations, the feature construction and decomposition module 220 includes: An exogenous feature construction submodule is used to construct exogenous features based on the target electricity sequence and the exogenous influence data; The multi-mode decomposition submodule is used to perform multi-mode feature decomposition on the target electrical quantity sequence to obtain multi-scale sequence components.

[0069] In some possible implementations, the exogenous feature construction submodule includes: The core feature unit is used to take the electricity consumption data in the electricity data as the core interpretation feature; The Spring Festival feature unit is used to construct a continuously quantified Spring Festival intensity feature based on the actual number of days the Spring Festival covers in the corresponding month. Meteorological feature units are used to construct meteorological disturbance characteristics based on the number of days with abnormally high or low temperatures in a region; The feature integration unit is used to integrate the core explanatory features, the Spring Festival intensity features, and the meteorological disturbance features to obtain the exogenous features. The exogenous features are used in conjunction with the target electricity sequence to perform multi-mode feature decomposition and component prediction on the target electricity sequence.

[0070] In some possible implementations, the multi-mode decomposition submodule includes: The X13 decomposition unit is used to decompose the target electrical quantity sequence using the X13 seasonal adjustment decomposition method to obtain trend cyclic components, seasonal components, and irregular components. The wavelet decomposition unit is used to decompose the target electrical quantity sequence by discrete wavelet decomposition to obtain low-frequency trend components and high-frequency detail components. The mode decomposition unit is used to adaptively decompose the target electrical quantity sequence using empirical mode decomposition to obtain intrinsic mode function components and residual trend components. The decomposition and integration unit is used to integrate the above decomposition results to obtain the multi-scale sequence components.

[0071] In some possible implementations, the component prediction and synthesis module 230 includes: The component prediction submodule is used to perform component prediction by matching the corresponding prediction model based on the features of each multi-scale sequence component. The result synthesis submodule is used to synthesize the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result.

[0072] In some possible implementations, the component prediction submodule includes: The linear prediction unit is used to predict the components of a linearly stationary multi-scale sequence using an autoregressive distributed lag model. The seasonal forecasting unit is used to forecast multi-scale sequence components with seasonal periodicity using a seasonal time series model with exogenous variables. The nonlinear prediction unit is used to predict the components of a nonlinear high-frequency multi-scale sequence using a long short-term memory network model. The prediction result output unit is used to obtain the component prediction results corresponding to each of the multi-scale sequence components through the above-mentioned targeted prediction processing, and send the component prediction results to the result synthesis submodule.

[0073] In some possible implementations, the evaluation and selection module 240 includes: The rolling forecasting submodule is used to perform rolling forecasts on an annual cycle. In each round of rolling forecasting, the actual electricity consumption data of the latest month is updated to the training data. The error calculation submodule is used to calculate the annual electricity sales forecast error corresponding to each round of rolling forecast; The compliance model filtering submodule is used to filter out prediction results in which the error of all rounds meets the preset constraints. The model selection submodule is used to select the model with the smallest average error among compliant models to complete the selection and obtain the final electricity sales forecast result.

[0074] The electricity sales forecasting device based on multi-feature decomposition provided in this invention can execute the electricity sales forecasting method based on multi-feature decomposition provided in any embodiment of this invention, and has the corresponding functional modules and beneficial effects of the method.

[0075] Figure 3 This is a schematic diagram of an electronic device for implementing the multi-feature decomposition-based electricity sales forecasting method according to embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0076] like Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0077] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0078] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a power sales forecasting method based on multi-feature decomposition.

[0079] In some embodiments, the electricity sales forecasting method based on multi-feature decomposition can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the electricity sales forecasting method based on multi-feature decomposition described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the electricity sales forecasting method based on multi-feature decomposition by any other suitable means (e.g., by means of firmware).

[0080] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0081] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0082] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0083] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0084] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0085] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0086] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0087] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for predicting electricity sales based on multi-feature decomposition, characterized in that, include: Acquire the power data and exogenous influence data of the target area, and preprocess the power data and exogenous influence data to obtain a standardized target power sequence; Based on the target power sequence and the exogenous influence data, exogenous features are constructed, and the target power sequence is decomposed in multiple ways to obtain multi-scale sequence components. Based on the feature matching of each multi-scale sequence component, the corresponding prediction model is used to predict the components, and the prediction results of each component are obtained. Then, the prediction results of each component are synthesized according to the decomposition rules to obtain the initial electricity sales prediction result. The initial electricity sales forecast results are evaluated and selected based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

2. The method according to claim 1, characterized in that, The preprocessing of the electricity data and exogenous influence data to obtain a standardized target electricity sequence includes: Convert the cumulative electricity sales and consumption data into the corresponding data for the current month; In the event of missing electricity data, missing value imputation is performed based on the electricity consumption characteristics of the corresponding missing month. All processed electricity data are standardized to the unit of ten thousand kilowatt-hours to obtain the standardized target electricity sequence.

3. The method according to claim 2, characterized in that, In the event of missing electricity data, the missing value imputation process is performed based on the electricity consumption characteristics of the corresponding missing month, including: In cases where data for the beginning of the year is missing, data imputation is performed based on the percentage of working days in the corresponding month. In the case of missing data for the last month of the year, regression interpolation or equivalent interpolation is performed based on the historical electricity consumption patterns of the same period.

4. The method according to claim 1, characterized in that, The construction of exogenous features based on the target electricity sequence and the exogenous influence data includes: The electricity consumption data in the electricity data is used as the core interpretation feature; Based on the actual number of days the Spring Festival covers in the corresponding month, a continuous quantitative feature of the intensity of the Spring Festival is constructed. Meteorological disturbance characteristics are constructed based on the number of days with abnormally high and low temperatures in the region; By integrating the core explanatory features, the Spring Festival intensity features, and the meteorological disturbance features, the exogenous features are obtained. The exogenous features, in conjunction with the target energy sequence, are used to perform multi-mode feature decomposition and component prediction on the target energy sequence.

5. The method according to claim 1, characterized in that, The step of performing multi-mode feature decomposition on the target energy sequence to obtain multi-scale sequence components includes: The target electricity sequence is decomposed using the X13 seasonal adjustment decomposition method to obtain trend cyclic components, seasonal components, and irregular components. The target electrical quantity sequence is decomposed using discrete wavelet decomposition to obtain low-frequency trend components and high-frequency detail components. The target electrical quantity sequence is adaptively decomposed using empirical mode decomposition to obtain intrinsic mode function components and residual trend components; By integrating the above decomposition results, the multi-scale sequence components are obtained.

6. The method according to claim 1, characterized in that, The step of performing component prediction based on feature matching of each multi-scale sequence component with a corresponding prediction model includes: For linear stationary multi-scale sequence components, an autoregressive distributed lag model is used for component prediction. For multi-scale sequence components with seasonal periodicity, a seasonal time series model with exogenous variables is used for component prediction. For nonlinear high-frequency multi-scale sequence components, a long short-term memory network model is used for component prediction. Through the above targeted prediction processing, the component prediction results corresponding to each of the multi-scale sequence components are obtained. The component prediction results are then synthesized according to the decomposition rules to obtain the initial electricity sales prediction result.

7. The method according to claim 1, characterized in that, The evaluation and selection of the initial electricity sales forecast results based on the rolling forecast mechanism to obtain the final electricity sales forecast results includes: Rolling forecasts are performed on an annual basis, and the training data is updated with the latest monthly actual electricity consumption data in each round of rolling forecasts. Calculate the annual electricity sales forecast error corresponding to each round of rolling forecasts; Filter out the prediction results where the error of all rounds meets the preset constraints; The model with the smallest average error is selected from the compliant models to complete the model selection, and the final electricity sales forecast result is obtained.

8. A power sales forecasting device based on multi-feature decomposition, characterized in that, include: The data acquisition and preprocessing module is used to acquire the power data and exogenous influence data of the target area, and preprocess the power data and exogenous influence data to obtain a standardized target power sequence. The feature construction and decomposition module is used to construct exogenous features based on the target electricity sequence and the exogenous influence data, and to perform multi-mode feature decomposition on the target electricity sequence to obtain multi-scale sequence components; The component prediction and synthesis module is used to perform component prediction based on the feature matching of each multi-scale sequence component with the corresponding prediction model, obtain the prediction results of each component, and synthesize the prediction results of each component according to the decomposition rules to obtain the initial electricity sales prediction result. The evaluation and selection module is used to evaluate and select the initial electricity sales forecast results based on the rolling forecast mechanism to obtain the final electricity sales forecast results.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the electricity sales forecasting method based on multi-feature decomposition as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the electricity sales forecasting method based on multi-feature decomposition as described in any one of claims 1-7.

Images