CN-122021994-A - Electric automobile charging demand prediction and dredging system based on holiday trip

CN122021994ACN 122021994 ACN122021994 ACN 122021994ACN-122021994-A

Abstract

The invention relates to the technical field of electric automobile charging, and discloses an electric automobile charging demand prediction and dredging system based on holiday travel, which comprises the steps of firstly constructing a multi-element heterogeneous data set integrating historical charging orders, high-speed passing policies and meteorological and road network construction information, and extracting specific characteristic factors of three holiday scenes of returning type, travel type and short distance type through cluster analysis and pattern recognition; and finally generating a multistage collaborative dispersion strategy of dynamic time-sharing electricity price of a demand side, pre-dispatching of an emergency charging cabin of a supply side, reverse drainage of a service area, dynamic flexible power distribution and queuing correction of a video blind area. The invention obviously improves the service efficiency of the holiday charging network and adapts to the dynamic fluctuation rules of different holiday scenes.

Inventors

ZHANG MINGKAI
DONG DELONG
WANG ZONGLIAN
YAO YUAN
CHEN LIANGLIANG
WANG WEI
REN BOQIANG
YOU XIAOYU
SI GANG
WANG GUOYU
CHEN JIAYU
ZHU HONGDONG
YANG ZHIHUA
ZHAO YINGCHUN
GAO SHUAI
LUN XIAOXIANG

Assignees

国网电动汽车服务(天津)有限公司
国网天津市电力公司
国家电网有限公司
北京南瑞普瑞用电技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251219

Claims (10)

1. The electric automobile charging demand prediction and drainage system based on holiday travel is characterized by comprising a data acquisition and feature extraction module, a charging demand prediction module and a multi-stage collaborative drainage module; the data acquisition and feature extraction module is used for collecting the multiple heterogeneous data sets and extracting the holiday travel feature factors of the user; The charging demand prediction module is constructed based on a space-time multi-head attention mechanism network, and outputs the time-period load demand and queuing congestion index of each charging station after inputting the holiday trip characteristic factors of the user; The multistage cooperative grooming module generates a multistage cooperative grooming strategy based on the time-of-day load demand and queuing congestion index of each charging station, and the multistage cooperative grooming strategy comprises a dynamic time-of-day electricity price mechanism, an emergency charging cabin time-space pre-dispatching model, a service area culvert opening and reverse drainage instruction, a dynamic flexible power distribution scheme and a vehicle access time sequence difference algorithm for correcting the queuing state in real time; The system realizes the improvement of the service efficiency of the charging network in holiday peak period through the synergistic effect of price guidance, physical space borrowing, power load flexible control and vision blind area compensation.
2. The electric vehicle charging demand prediction and dispersion system based on holiday travel according to claim 1, wherein the multi-component heterogeneous data set comprises historical charging order data, high-speed passing policy information, meteorological environment data and road network construction information, and the user holiday travel characteristic factors cover exclusive travel mode characteristics in returning, traveling and short-distance scenes.
3. The electric vehicle charging demand prediction and drainage system based on holiday travel of claim 1, wherein the data acquisition and feature extraction module comprises a multi-element data acquisition unit and a feature factor extraction unit, The multi-element data acquisition unit acquires corresponding data through a charging pile platform, a traffic management department, a weather bureau and a road maintenance department respectively, performs time sequence alignment and noise filtering through a Kalman filtering algorithm, recognizes and eliminates abnormal values through an isolated forest algorithm to form a multi-element heterogeneous data set in a standardized tensor form, and the data acquisition frequency is real-time or near real-time; The feature factor extraction unit divides holiday scenes into three types of countryside, travel and short-distance types based on cluster analysis, and extracts the countryside tidal intensity factor, the travel divergence factor and the short-distance burst factor respectively through Principal Component Analysis (PCA), linear Discriminant Analysis (LDA) and STL) time sequence decomposition technologies to form key feature factor vectors.
4. The electric vehicle charging demand prediction and drainage system based on holiday travel of claim 3, wherein the tidal intensity factor calculation formula is: ; Wherein, the And High-speed inlet and outlet flows, respectively; the divergence factor calculation formula is as follows: ; Wherein, the For the distance of the ith station to the attraction, Is the average distance; The burst factor calculation formula is as follows: ; Wherein, the And Peak and base load, respectively.
5. The electric vehicle charging demand prediction and drainage system based on holiday travel of claim 1, wherein the charging demand prediction module comprises an ST-MHA network architecture, an encoder-decoder structure, a Kalman filter correction unit and a multi-task learning framework, The ST-MHA network architecture is derived from a transducer architecture and improved, is divided into a space attention sub-module and a time attention sub-module, calculates space correlation features and time correlation features respectively through a multi-head attention mechanism, and fuses space-time features through splicing operation; The encoder consists of a plurality of ST-MHA layers and a feedforward neural network FFN, each sub-layer adopts residual connection and layer normalization, and the decoder outputs a prediction result through two ST-MHA layers and Softmax functions; the Kalman filtering correction unit is used for calibrating the predicted value in real time through a state equation, a measurement equation and a correction equation; the multi-task learning framework is characterized in that a loss function is a weighted sum of MSE loss of load prediction and cross entropy loss of queuing congestion index, and two prediction tasks are optimized synchronously.
6. The holiday trip-based electric vehicle charging demand prediction and drainage system according to claim 5, wherein the queuing congestion index calculation formula is as follows: ; Wherein, the In order to line up the number of vehicles, Maximum capacity for a site; The calculation formula of the optimal state estimation value of the Kalman filter correction unit is as follows: ; Wherein, the To predict the value at time t from time t-1, For the Kalman gain, H is the measurement matrix.
7. The electric vehicle charging demand prediction and dispersion system based on holiday travel of claim 1, wherein the multi-stage cooperative dispersion module comprises a demand side guiding unit, a supply side scheduling unit and a queuing state correcting unit, The demand side guiding unit implements a dynamic time-of-use electricity price mechanism, and the electricity price is obtained by the basic electricity price Dynamic service fees The service charge is adjusted in real time along with the queuing congestion index, and the electricity price formula is as follows: ; wherein k is the price sensitivity coefficient, In order to be able to queue the congestion index, Is a congestion threshold; The supply side dispatching unit comprises an emergency charging cabin time-space pre-dispatching model, a service area culvert opening and reverse drainage model and a dynamic flexible power distribution model, The emergency charging cabin time-space pre-dispatching model optimizes the throwing quantity and the throwing positions through linear programming based on load gap prediction, and adopts Calculating an optimal transportation path by an algorithm; Triggering culvert opening instructions when the up-down load unbalance degree exceeds a set value by the service area culvert opening and reverse drainage model; The dynamic flexible power distribution model reduces the output power of a single gun to a flexible value when the predicted load exceeds a transformer threshold value; the queuing state correction unit calculates the inlet flow through a vehicle inlet and outlet time sequence difference algorithm And outlet flow rate And (3) the difference value of the hidden queuing vehicles is estimated by fusion of Kalman filtering.
8. The electric vehicle charging demand prediction and drainage system based on holiday travel of claim 5, wherein position codes are introduced into the ST-MHA network architecture, time sequence information is coded through sine and cosine functions of different frequencies, and a position coding formula is as follows: ; Wherein, the For the position index to be used, Is the dimension of the model vector.
9. The electric vehicle charging demand prediction and drainage system based on holiday travel according to claim 1, wherein the vehicle in-out time sequence difference algorithm processes a flow difference by adopting a sliding window average filter, the window size is set to be 5 minutes, and the number of vehicles in hidden queuing is calculated by integration: ; Wherein, the In order to provide a sampling interval, For the number of vehicles to be queued implicitly, And triggering a precise induction strategy when the number of the vehicles in the hidden queuing exceeds a dynamic threshold value for the average value of the traffic flow difference at the entrance and the exit.
10. The electric automobile charging demand prediction and dispersion system based on holiday travel, which is disclosed in claim 1, is characterized by further comprising a user interaction module, wherein the user interaction module is used for providing real-time charging suggestions, electricity price information and queuing states for users through mobile APP and vehicle-mounted terminals, and the interaction module is used for generating a personalized scheme based on user preferences so as to improve user experience.

Description

Electric automobile charging demand prediction and dredging system based on holiday trip Technical Field The invention belongs to the technical field of electric vehicle charging, and particularly relates to an electric vehicle charging demand prediction and drainage system based on holiday travel. Background Along with the rapid promotion of electric automobile popularity, user concentrated trip leads to the charging demand to appear the explosive growth during holiday, and charging network faces serious challenge: Firstly, the space-time distribution of the charging demands is extremely uneven, the tidal flow characteristics under the scenes of spring festival return to village, national celebration travel and the like are obvious, and an unbalanced state of 'one side is jammed and one side is idle' is often generated in a high-speed service area; Secondly, the existing charging demand prediction technology relies on simple time sequence models such as ARIMA and the like to predict based on historical load data only, and lacks suitability for special scenes (such as return country, travel and short trip) of holidays, so that the prediction precision is insufficient; Thirdly, the dredging strategy mostly adopts fixed time-of-use electricity price or static resource scheduling, can not respond to the fluctuation of demand in real time, and is difficult to solve the problems of 'pile-on-no-electricity' and 'hidden queuing' caused by video monitoring blind areas due to the limited capacity of the transformer; fourth, the multisource data fusion is insufficient, and influences of external factors such as high-speed traffic policies, meteorological conditions and road network construction on traveling and charging behaviors of users are not fully considered, so that robustness of a prediction model is poor. The prior art has obvious defects in aspects of scene adaptation, multi-source data cooperative utilization, dynamic accurate dredging and the like of holiday charging requirements, and cannot meet the requirement of efficient operation of a charging network, so that an integrated technical scheme for fusing scene characteristics, high-precision prediction and multi-dimensional cooperative dredging is needed. Disclosure of Invention The invention aims to overcome the defects in the prior art and provides an electric vehicle charging demand prediction and dredging system based on holiday travel. The invention solves the technical problems by the following technical proposal: the electric automobile charging demand prediction and drainage system based on holiday travel is characterized by comprising a data acquisition and feature extraction module, a charging demand prediction module and a multi-stage collaborative drainage module; the data acquisition and feature extraction module is used for collecting the multiple heterogeneous data sets and extracting the holiday travel feature factors of the user; The charging demand prediction module is constructed based on a space-time multi-head attention mechanism network, and outputs the time-period load demand and queuing congestion index of each charging station after inputting the holiday trip characteristic factors of the user; The multistage cooperative grooming module generates a multistage cooperative grooming strategy based on the time-of-day load demand and queuing congestion index of each charging station, and the multistage cooperative grooming strategy comprises a dynamic time-of-day electricity price mechanism, an emergency charging cabin time-space pre-dispatching model, a service area culvert opening and reverse drainage instruction, a dynamic flexible power distribution scheme and a vehicle access time sequence difference algorithm for correcting the queuing state in real time; The system realizes the improvement of the service efficiency of the charging network in holiday peak period through the synergistic effect of price guidance, physical space borrowing, power load flexible control and vision blind area compensation. And the multi-element heterogeneous data set comprises historical charging order data, high-speed passing policy information, meteorological environment data and road network construction information, and the holiday trip characteristic factors of the user cover exclusive trip mode characteristics in returning type, traveling type and short-distance scenes. Moreover, the data acquisition and feature extraction module comprises a multi-element data acquisition unit and a feature factor extraction unit, The multi-element data acquisition unit acquires corresponding data through a charging pile platform, a traffic management department, a weather bureau and a road maintenance department respectively, performs time sequence alignment and noise filtering through a Kalman filtering algorithm, recognizes and eliminates abnormal values through an isolated forest algorithm to form a multi-element heterogeneous data set in a standardized tenso