CN-121983967-A - Charging load prediction method and device based on time sequence decomposition and electronic equipment

Abstract

The invention discloses a charging load prediction method and device based on time sequence decomposition and electronic equipment, belongs to the technical field of load prediction, and aims to solve the problem that a traditional prediction method is difficult to effectively capture a complex seasonal rule of charging an electric automobile. The method comprises the steps of obtaining historical load data, wherein the historical load data comprises load values and meteorological environment information in a preset period, obtaining prior features and basic features based on the historical load data, wherein the basic features comprise at least one of time index features and meteorological environment features, the prior features are obtained by decomposing time sequence features of the historical load data through a Prophet model, fusing the prior features and the basic features to obtain input feature vectors, and inputting the input feature vectors into a pre-trained prediction module to conduct load prediction to obtain a prediction result. The embodiment of the invention can improve the accuracy of the charging load prediction result.

Inventors

• LIU CHUCHU
• LU XIAOXUAN
• LI LUQI

Assignees

• 长沙理工大学

Dates

Publication Date

20260505

Application Date

20260330

Claims (10)

1. 1. A charging load prediction method based on time sequence decomposition is characterized by comprising the following steps: acquiring historical load data, wherein the historical load data comprises load values and weather environment information in a preset period; Acquiring prior features and basic features based on the historical load data, wherein the basic features comprise time index features and meteorological environment features, the prior features are obtained by decomposing time sequence features of the historical load data by a Prophet model, the time index features are determined based on time information in the preset period, and the meteorological environment features are determined based on the meteorological environment information; fusing the prior feature and the basic feature to obtain an input feature vector; And inputting the input feature vector into a pre-trained prediction module to perform load prediction, so as to obtain a prediction result.
2. 2. The time-series decomposition-based charge load prediction method according to claim 1, wherein the obtaining a priori features based on the historical load data includes: Decomposing the load value into a combination of trend terms, seasonal terms, holiday terms and error terms by a propset model, wherein the trend terms are used for fitting aperiodic baseline drift, the seasonal terms are used for fitting periodic fluctuation of the load value on a weekly and annual scale, the holiday terms are used for capturing load mutation during legal holidays, and the error terms are random errors obeying normal distribution; Extracting a value of the trend term at each time as a trend component, extracting a value of the seasonal term at each time as a seasonal component, extracting a value of the holiday term at each time as a holiday component, and generating a predicted value component based on the trend term, the seasonal term, and the holiday term; An a priori feature is determined, the a priori feature including at least one of the trend component, the seasonal component, the holiday component, and the predictor component.
3. 3. The charging load prediction method based on time-series decomposition according to claim 2, wherein the seasonal component includes a daily cycle characteristic component and a weekly cycle characteristic component, and the extracting the value of the seasonal item at each time as the seasonal component includes: performing multi-scale decomposition on the seasonal items, and respectively extracting the daily cycle characteristic component and the weekly cycle characteristic component; The daily cycle characteristic component is used for fitting load waveforms of a morning peak, a evening peak and a night valley based on a 24-hour Fourier series, and the weekly cycle characteristic component is used for fitting load differences of a working day and a weekend based on a 7-day Fourier series.
4. 4. The charging load prediction method based on time series decomposition according to claim 2, wherein the time of day is determined by a propset model Load value of (2) The combination of trend term, seasonal term, holiday term and error term is as follows: ; Or, the time of day is calculated by a Prophet model Load value of (2) The combination of trend term, seasonal term, holiday term and error term is as follows: ; Wherein, the As a result of the trend term(s), For the seasonal term in question, For the holiday term in question, Is the error term.
5. 5. The time-series decomposition-based charge load prediction method of claim 1, wherein the prediction module is constructed based on an extreme gradient lifting XGBoost model.
6. 6. The charging load prediction method based on time-series decomposition according to claim 1, wherein the basic features further include at least one of: A short-term hysteresis characteristic comprising load values from 1 st to nth hours prior to the current time instant for characterizing a short-term autocorrelation of the load, N being a positive integer; the cycle hysteresis feature comprises load values at corresponding moments of the first M cycles with the same cycle phase as the current moment, and is used for capturing a cycle dependence rule of the load on a daily scale or a week scale, wherein M is a positive integer; And rolling statistical characteristics calculated based on a plurality of preset time windows, wherein the rolling statistical characteristics comprise at least one of rolling mean values, rolling maximum values and rolling standard deviations of a load sequence in the preset time windows and are used for representing the concentration trend and the discrete degree of the load in a local time range.
7. 7. The time-series decomposition-based charge load prediction method of claim 1, wherein the weather-environment information includes temperature, humidity, rainfall and surface level radiation, and the weather-environment characteristics include at least one of: The basic weather features comprise temperature features corresponding to the temperature, humidity features corresponding to the humidity, rainfall features corresponding to the rainfall and radiation features corresponding to the surface horizontal radiation; A first boolean feature for characterizing whether it is high temperature weather; a second boolean feature for characterizing whether or not to rainfall; and the second-order interaction characteristic is obtained by multiplying the normalized basic meteorological characteristic by two points.
8. 8. A charging load prediction apparatus based on time-series decomposition, comprising: the first acquisition module is used for acquiring historical load data, wherein the historical load data comprises load values and weather environment information in a preset period; The second acquisition module is used for acquiring prior characteristics and basic characteristics based on the historical load data, wherein the basic characteristics comprise time index characteristics and meteorological environment characteristics, the prior characteristics are obtained by decomposing the time sequence characteristics of the historical load data by a Prophet model, the time index characteristics are determined based on time information in the preset period, and the meteorological environment characteristics are determined based on the meteorological environment information; the fusion module is used for fusing the priori features and the basic features to obtain input feature vectors; And the input module is used for inputting the input feature vector into the pre-trained prediction module to perform load prediction, so as to obtain a prediction result.
9. 9. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, characterized in that the processor is adapted to read the program in the memory to implement the steps in the charging load prediction method based on time-series decomposition according to any one of claims 1 to 7.
10. 10. A readable storage medium storing a program, wherein the program when executed by a processor implements the steps in the time-series decomposition-based charge load prediction method according to any one of claims 1 to 7.

Description

Charging load prediction method and device based on time sequence decomposition and electronic equipment Technical Field The invention belongs to the technical field of load prediction, and particularly relates to a charging load prediction method and device based on time sequence decomposition and electronic equipment. Background The rapid popularization of new energy automobiles brings new electricity load increasing points. This growth drives a rapid rise in the charging load of the electric vehicle. Such a large-scale access results in a charging load exhibiting a high degree of uncertainty and coupling in both the time and space dimensions. From a microscopic mechanism, the charging requirement of the electric automobile is not only limited by the travel rules of users (such as commuter tides and holiday travel), but also shows complex nonlinear association with the meteorological environment. For example, the extreme high and low temperatures may significantly change the charge and discharge characteristics of the power battery and the energy consumption of the air conditioner in the vehicle, thereby causing distortion of the load curve. Furthermore, different types of charging stations (e.g., residential slow-fill, high-speed fast-fill stations) have distinct daily load profiles. The correlation and heterogeneity of this multifactor interleaving makes deep periodic laws and abrupt features in charge-loaded sequences that are difficult to capture by traditional linear models implicit. How to construct effective characteristic variables and to mine the internal rule of the charging load of the electric automobile becomes a key for improving the prediction precision. The prediction method in the prior art is difficult to effectively capture the problem of complex seasonal regularity of electric automobile charging, so that the accuracy of a charging load prediction result is low. Disclosure of Invention The invention aims to solve the technical problem that the method commonly used in the prior art is difficult to effectively capture the complicated seasonal rule of electric automobile charging, so that the charging meets the problem of low accuracy of a prediction result. The content of the invention comprises: in a first aspect, an embodiment of the present invention provides a charging load prediction method based on time-series decomposition, including: acquiring historical load data, wherein the historical load data comprises load values and weather environment information in a preset period; Acquiring prior features and basic features based on the historical load data, wherein the basic features comprise time index features and meteorological environment features, the prior features are obtained by decomposing time sequence features of the historical load data by a Prophet model, the time index features are determined based on time information in the preset period, and the meteorological environment features are determined based on the meteorological environment information; fusing the prior feature and the basic feature to obtain an input feature vector; And inputting the input feature vector into a pre-trained prediction module to perform load prediction, so as to obtain a prediction result. Optionally, the acquiring a priori features based on the historical load data includes: Decomposing the load value into a combination of trend terms, seasonal terms, holiday terms and error terms by a propset model, wherein the trend terms are used for fitting aperiodic baseline drift, the seasonal terms are used for fitting periodic fluctuation of the load value on a weekly and annual scale, the holiday terms are used for capturing load mutation during legal holidays, and the error terms are random errors obeying normal distribution; Extracting a value of the trend term at each time as a trend component, extracting a value of the seasonal term at each time as a seasonal component, extracting a value of the holiday term at each time as a holiday component, and generating a predicted value component based on the trend term, the seasonal term, and the holiday term; An a priori feature is determined, the a priori feature including at least one of the trend component, the seasonal component, the holiday component, and the predictor component. Optionally, the seasonal component includes a daily cycle characteristic component and a weekly cycle characteristic component, and the extracting the value of the seasonal item at each time as the seasonal component includes: performing multi-scale decomposition on the seasonal items, and respectively extracting the daily cycle characteristic component and the weekly cycle characteristic component; The daily cycle characteristic component is used for fitting load waveforms of a morning peak, a evening peak and a night valley based on a 24-hour Fourier series, and the weekly cycle characteristic component is used for fitting load differences of a working day and a weekend b