CN-122021986-A - Space-time feature fusion-based electricity consumption prediction method and system

CN122021986ACN 122021986 ACN122021986 ACN 122021986ACN-122021986-A

Abstract

The application discloses a power consumption prediction method and system based on space-time feature fusion, relates to the technical field of power consumption prediction, and solves the problem of insufficient prediction precision in power consumption prediction. According to the embodiment of the application, the space-time characteristics are constructed by introducing the target external data with causal relation or statistical correlation with the electricity consumption, the corresponding factor prediction model and the CNN-LSTM prediction model are constructed and trained, the future external factors can be predicted first, and then the future electricity consumption can be predicted based on the future external factors and the historical electricity consumption data. The method of the embodiment of the application can improve the comprehensiveness and the prediction precision of prediction.

Inventors

ZHANG YONGYIN
Zhong Yusha
WANG ZHICHENG
XIAO HAIHUA
XU FENGYANG
TAN KE

Assignees

广东电网有限责任公司管理科学研究院

Dates

Publication Date: 20260512
Application Date: 20251208

Claims (10)

1. The power consumption prediction method based on space-time feature fusion is characterized by comprising the following steps of: acquiring historical electricity consumption data and target external data, wherein the target external data is generated outside the power system and has causal relation or statistical correlation with the electricity consumption; Constructing a space-time feature based on the historical electricity consumption data and the target external data, wherein the space-time feature of each time point comprises the target external data of the corresponding time point and the historical electricity consumption data before the corresponding time point; Constructing a corresponding factor prediction model for different types of external factors in the target external data, wherein the factor prediction model is used for predicting a factor prediction value in a future preset period based on historical data of the corresponding external factors; Based on the historical electricity consumption data and the space-time characteristics, an improved convolutional neural network CNN-long-short-term memory LSTM prediction model is constructed and trained; and predicting the electricity consumption in the future preset period based on the factor prediction model and the CNN-LSTM prediction model.
2. The method of claim 1, wherein the constructing a spatiotemporal feature based on the historical power usage data and the target external data comprises: splitting the historical electricity consumption data into a long-term trend component and a seasonal fluctuation component through a time sequence decomposition algorithm; Extracting long-term trend characteristics of the long-term trend component through a trend extraction algorithm; extracting the seasonal wave characteristics of the seasonal wave components by a spectrum analysis method; Respectively carrying out data dimension reduction on the power consumption characteristics and the characteristics of the target external data by a principal component analysis method to obtain dimension-reduced power consumption characteristics and dimension-reduced external characteristics, wherein the power consumption characteristics comprise the long-term trend characteristics and the seasonal fluctuation characteristics; and fusing the dimension-reducing electricity consumption characteristic and the dimension-reducing external characteristic to obtain the space-time characteristic.
3. The method of claim 2, wherein the target external data corresponding to the dimension reduction external feature includes a domestic total production value GDP, an air temperature, and a month work day.
4. The method of claim 3, wherein the constructing a corresponding factor prediction model for different types of external factors in the target external data comprises: constructing an ARIMA time sequence prediction model aiming at the fluctuation characteristic of the GDP; Constructing a seasonal decomposition prediction model according to the change rule of the air temperature; And carrying out deterministic modeling on the basis of an calendar rule aiming at the month working days.
5. The method of claim 1, wherein after said constructing the corresponding factor prediction model, the method further comprises: Determining a scene boundary of the external factor by adopting a fractional method based on the target external data, wherein the scene boundary is used for dividing a high factor value scene, a medium factor value scene and a low factor value scene corresponding to the external factor; Randomly sampling under the constraint of the scene boundary through Monte Carlo simulation, and generating a factor prediction value of a preset quantity group for each scene based on the factor prediction model and sampling data; counting the mean value and the fluctuation interval of the factor predicted value under each scene; The predicting the electricity consumption in the future preset period based on the external factor prediction model and the CNN-LSTM prediction model comprises the following steps: constructing scene space-time characteristics based on the average value and fluctuation interval of the factor predicted values under each scene and the historical electricity consumption data; And inputting the scene space-time characteristics into the CNN-LSTM prediction model to obtain the probability distribution of the electricity consumption in the future preset period output by the CNN-LSTM prediction model.
6. The method of claim 5, wherein constructing scene spatiotemporal features based on the mean and fluctuation interval of the factor predictors for each scene and the historical electricity usage data comprises: constructing a reference scene space-time characteristic based on the average value of the factor predicted values under each scene and the characteristic of the historical electricity consumption data; Randomly generating fluctuation amplitude characteristics based on the fluctuation interval; and constructing the scene space-time characteristic based on the reference scene space-time characteristic and the fluctuation amplitude characteristic.
7. A power consumption prediction system based on space-time feature fusion, applied to the method of any one of claims 1-6, the system comprising: the system comprises an acquisition module, a storage module and a storage module, wherein the acquisition module is used for acquiring historical electricity consumption data and target external data, and the target external data is generated outside an electric power system and has causal relation or statistical correlation with the electricity consumption; The characteristic engineering module is used for constructing space-time characteristics based on the historical electricity consumption data and the target external data, wherein the space-time characteristics of each time point comprise the target external data of the corresponding time point and the historical electricity consumption data before the corresponding time point; The system comprises a target external data, a first modeling module, a second modeling module and a first modeling module, wherein the target external data comprises a target external data, the first modeling module is used for constructing a corresponding factor prediction model for different types of external factors in the target external data, and the factor prediction model is used for predicting a factor prediction value in a future preset period based on historical data of the corresponding external factors; The second modeling module is used for constructing and training an improved convolutional neural network CNN-long-short-term memory LSTM prediction model based on the historical electricity consumption data and the space-time characteristics; And the prediction module is used for predicting the electricity consumption in the future preset period based on the factor prediction model and the CNN-LSTM prediction model.
8. The system of claim 7, further comprising a simulation module configured to determine a scenario boundary of the external factor using a quantile method based on the target external data, wherein the scenario boundary is configured to divide a high factor value scenario, a medium factor value scenario, and a low factor value scenario corresponding to the external factor; The prediction module is specifically used for constructing scene space-time characteristics based on the mean value and the fluctuation interval of the factor predicted values under each scene and the historical electricity consumption data, inputting the scene space-time characteristics into the CNN-LSTM prediction model, and obtaining the probability distribution of the electricity consumption in the future preset period output by the CNN-LSTM prediction model.
9. A computing device, comprising: A memory for storing a program; A processor for loading the program to perform the method of any of claims 1-6.
10. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored program, wherein the program, when run, controls a device in which the computer readable storage medium is located to perform the method of any one of claims 1-6.

Description

Space-time feature fusion-based electricity consumption prediction method and system Technical Field The invention relates to the technical field of electricity consumption prediction, in particular to an electricity consumption prediction method and system based on space-time feature fusion. Background In the operation and management process of the power system, the power consumption prediction is a core technical link for guaranteeing the safe and stable operation of the power grid, realizing the optimal configuration of energy resources and improving the accuracy of business decisions of power enterprises. The accurate electricity consumption prediction can provide scientific basis for the power grid dispatching department, help the power grid dispatching department to reasonably arrange the start-stop plan of the generator set and coordinate the trans-regional power transmission, thereby effectively avoiding power grid fluctuation or energy waste caused by unbalanced power supply and demand, and simultaneously, can provide data support for business decisions such as power market transaction, power facility planning and the like, and has important significance for improving the overall operation efficiency and economic benefit of a power system. However, the traditional electricity consumption prediction method is mainly based on data analysis of a single dimension, lacks comprehensive consideration of multi-dimension factors influencing electricity consumption change, and is difficult to capture potential association between the multi-dimension influencing factors and the electricity consumption, so that prediction accuracy is insufficient. In view of this, there is a need for a method and system for power consumption prediction based on temporal-spatial feature fusion. Disclosure of Invention Aiming at the problem of insufficient prediction precision in the process of predicting the power consumption in the prior art, the invention provides a power consumption prediction method and a power consumption prediction system based on space-time feature fusion, which can improve the prediction precision of power consumption prediction. The specific technical scheme is as follows: In a first aspect, an embodiment of the present application provides a power consumption prediction method based on space-time feature fusion, including: The method comprises the steps of obtaining historical electricity consumption data and target external data, wherein the target external data are data which are generated outside an electric power system and have causal relation or statistical correlation with electricity consumption, constructing space-time characteristics based on the historical electricity consumption data and the target external data, wherein the space-time characteristics of each time point comprise the target external data of the corresponding time point and the historical electricity consumption data before the corresponding time point, constructing corresponding factor prediction models for different types of external factors in the target external data, wherein the factor prediction models are used for predicting factor prediction values in a future preset period based on the historical data of the corresponding external factors, constructing and training an improved convolutional neural network CNN-long-short term memory LSTM prediction model based on the historical electricity consumption data and the space-time characteristics, and predicting the electricity consumption in the future preset period based on the factor prediction models and the CNN-LSTM prediction model. The method comprises the steps of dividing historical electricity consumption data into a long-term trend component and a seasonal fluctuation component through a time sequence decomposition algorithm, extracting the long-term trend characteristic of the long-term trend component through a trend extraction algorithm, extracting the seasonal fluctuation characteristic of the seasonal fluctuation component through a spectrum analysis method, respectively carrying out data dimension reduction on the electricity consumption characteristic and the characteristic of target external data through a principal component analysis method to obtain a dimension-reduced electricity consumption characteristic and a dimension-reduced external characteristic, wherein the electricity consumption characteristic comprises the long-term trend characteristic and the seasonal fluctuation characteristic, and fusing the dimension-reduced electricity consumption characteristic and the dimension-reduced external characteristic to obtain the space-time characteristic. Preferably, the target external data corresponding to the dimension reduction external feature includes a domestic total production value GDP, an air temperature and a month working days. Preferably, the construction of the corresponding factor prediction model for different types of external factors in