CN-121981262-A - Intelligent electric vehicle charging method and system based on reasoning

CN121981262ACN 121981262 ACN121981262 ACN 121981262ACN-121981262-A

Abstract

The invention discloses an intelligent charging method and system for an electric vehicle based on reasoning, which belong to the technical field of electric pile power regulation and control, and are used for predicting a charging scene by collecting relevant parameters of the electric vehicle during charging in real time and constructing a light-weight multi-task learning network, so that not only can the state of a power grid be dynamically perceived, but also the charging strategy can be dynamically regulated according to the state of the power grid, and a plurality of independent models are not required to be deployed through light-weight processing, so that the multi-task parallel processing capability is improved, the reasoning efficiency is improved, the charging efficiency is maximized on the premise of ensuring the safety of the power grid, the balance of the charging speed and the power grid protection is realized, and the problems that the existing electric pile power regulation and control mostly adopts static control logic, the state of the power grid is difficult to be dynamically perceived, and the output power is actively regulated according to the fluctuation trend of the power grid are solved, and the running efficiency of the charging pile is low are caused.

Inventors

WANG HAO
LIN DESHUN
WEN JIAXING
Gui Qinghong

Assignees

正泰安能数字能源(浙江)股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (10)

1. An electric vehicle intelligent charging method based on reasoning is characterized by comprising the following steps: S1, collecting electric power parameters and environment parameters of an electric vehicle during charging in real time, and preprocessing to obtain original data; S2, constructing a light multi-task learning network, dividing original data into corresponding data prediction tasks, and predicting a charging scene; And S3, optimizing the prediction result based on the multi-task loss optimization mechanism to obtain a target prediction result, and outputting an optimal charging strategy of the electric vehicle based on the target prediction result and combining with the dynamic regulation strategy.
2. The reasoning-based intelligent charging method for the electric vehicle according to claim 1, wherein the method comprises the following steps: the electric power parameters comprise charging voltage, charging current, active power, frequency, PWM duty ratio and power gear, and the environment parameters comprise weather and environment temperature.
3. The reasoning-based intelligent charging method for the electric vehicle according to claim 1, wherein the method comprises the following steps: s2, constructing a light multi-task learning network, dividing original data into corresponding data prediction tasks to predict a charging scene, and comprising the following steps: A one-dimensional convolutional neural network is used as a main body to construct a feature sharing layer and a specific task output layer to obtain a light multi-task learning network; the feature sharing layer comprises a convolution layer and a batch normalization layer, and replaces a full connection layer with a global maximum pooling layer, wherein the convolution layer is used for extracting features of original data to obtain the original feature data; normalizing the original feature data by a batch normalization layer to obtain training data, and obtaining a target feature vector by taking the maximum value of the corresponding type data in the training data by the global maximum pooling layer; and the specific task output layer divides the original data into corresponding data prediction tasks based on the target feature vector to predict the charging scene.
4. The reasoning-based intelligent charging method for the electric vehicle according to claim 3, wherein the method comprises the following steps: the data prediction task comprises a power grid state prediction task, a voltage trend prediction task and a power gear recommendation task; the power grid state prediction task outputs abnormal probability of each time node of the power grid in a prediction time domain in a two-class prediction value form to obtain a power grid state prediction value; the voltage trend prediction task outputs the variation of the voltage of the power grid in a prediction time domain in a linear regression mode to obtain a voltage trend prediction value; And the power gear recommending task outputs probability distribution of low, medium and high three-gear charging gears in the form of three classification output values to obtain the power recommended gears.
5. The reasoning-based intelligent charging method for the electric vehicle as claimed in claim 4, wherein: in S3, optimizing the predicted result based on a multi-task loss optimization mechanism to obtain a target predicted result, wherein the method comprises the following steps of: Calculating a binary cross entropy of the power grid state predicted value to obtain power grid state predicted deviation; calculating the mean square error of the voltage trend predicted value to obtain voltage trend predicted deviation; Calculating multi-class cross entropy of the power recommended gear to obtain power recommended deviation; Carrying out weighted summation on the power grid state prediction deviation, the voltage trend prediction deviation and the power recommendation deviation based on a preset loss allocation weight to obtain comprehensive prediction loss; and if the comprehensive prediction loss is greater than or equal to the loss threshold value, adjusting configuration parameters of the light-weight multi-task learning network, and carrying out prediction again until the comprehensive prediction loss is smaller than the loss threshold value to obtain a target prediction result.
6. The reasoning-based intelligent charging method for the electric vehicle according to claim 1, wherein the method comprises the following steps: S3, outputting an optimal charging strategy of the electric vehicle based on a target prediction result and combining with a dynamic regulation strategy, wherein the method comprises the following steps of: If the power grid state predicted value is smaller than the first abnormal threshold value and the maximum fluctuation of the voltage trend predicted value is in the first fluctuation range, gradually increasing the output power to a high gear; If the power grid state predicted value is between the first abnormal threshold value and the second abnormal threshold value, keeping the output power unchanged in the middle gear; If the power grid state predicted value is greater than or equal to a second abnormal threshold value and the maximum fluctuation of the voltage trend predicted value is in a second fluctuation range, gradually reducing the output power and limiting the highest output power to be a middle gear; And if the power grid state predicted value is greater than or equal to the second abnormal threshold value and the maximum fluctuation of the voltage trend predicted value is in the third fluctuation range, gradually reducing the output power and limiting the highest output power to be in a low gear.
7. The reasoning-based intelligent charging method of the electric vehicle as claimed in claim 6, wherein: And if the number of the charged electric vehicles is greater than or equal to a charging threshold value or the ambient temperature is greater than or equal to a temperature threshold value, limiting the highest output power to be a middle gear.
8. An electric vehicle intelligent charging system based on reasoning is applicable to the electric vehicle intelligent charging method based on reasoning as set forth in any one of claims 1-7, and is characterized by comprising a data acquisition module, a multi-task learning module, a loss optimization module and a strategy generation module; the data acquisition module acquires power parameters and environment parameters of the electric vehicle during charging in real time, and performs preprocessing to obtain original data; The multi-task learning module carries a light multi-task learning network, divides the original data into corresponding data prediction tasks and predicts a charging scene; The loss optimization module optimizes the prediction result based on a multi-task loss optimization mechanism to obtain a target prediction result; and the strategy generation module outputs the optimal charging strategy of the electric vehicle based on the target prediction result and in combination with the dynamic regulation strategy.
9. The computer equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus, the memory is used for storing a computer program, and the processor is used for realizing the steps of the intelligent charging method of the electric vehicle based on reasoning according to any one of claims 1-7 when executing the program stored on the memory.
10. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program, which when executed by a processor, implements the steps of an inference based intelligent charging method for electric vehicles according to any one of claims 1-7.

Description

Intelligent electric vehicle charging method and system based on reasoning Technical Field The invention relates to the technical field of electricity utilization regulation of charging piles, in particular to an intelligent charging method and system for an electric vehicle based on reasoning. Background With the popularization of urban two-wheeled electric vehicles, the demand for centralized charging piles is rapidly increasing. However, in the peak period of electricity consumption, a plurality of charging devices are operated concurrently, voltage fluctuation and even impact are easily generated on a local power grid, and the problems of system overload, frequent power protection action, low power distribution efficiency and the like are caused. The traditional charging pile mostly adopts fixed threshold control logic (such as voltage/current protection) or a passive current limiting strategy, cannot dynamically sense the state of a power grid, and cannot actively adjust power according to fluctuation trend, so that the system lacks self-adaptive capacity and adjustment flexibility to the power grid. Although the deep learning algorithm has achieved better results in the field of power prediction in recent years, most models depend on cloud computing power support and are not suitable for edge-side or embedded equipment. In application scenes such as a two-wheel vehicle charging pile, deployment hardware resources are limited, and how to locally run a deep learning model and realize real-time control is a current technical difficulty and development bottleneck. The Chinese patent publication No. CN116587916A, publication No. 2023, 8 and 15 discloses an intelligent charging method, a charging pile, computer equipment and a storage medium of an electric vehicle, wherein the state of charge of each electric vehicle at a plurality of time points is acquired through a detection unit, a first state vector of each charging cluster is calculated according to the state of charge of the electric vehicle and is uploaded to a second control unit, a reward value of each received first state vector is input into a neural network of the first control unit to obtain a first charging power distribution scheme of each second control unit, and a parameter set is repeatedly acquired and charging power distribution is carried out for a plurality of times until the variance of the state vector of each charging cluster is smaller than or equal to the preset value. Although the charging distribution scheme is selected by collecting the charge state of the electric vehicle, the charging loss can be reduced to a certain extent, if the charging distribution scheme is applied to application scenes such as two-wheel vehicle charging piles, the deployment hardware resources are limited, and the corresponding effect is difficult to be exerted. Disclosure of Invention Aiming at the problems that the existing charging pile power utilization regulation and control mostly adopts static control logic, the power grid state is difficult to be perceived dynamically, and the output power is regulated actively according to the fluctuation trend of the power grid, so that the operation efficiency of the charging pile is low, the invention provides an intelligent charging method and system for the electric vehicle based on reasoning. In a first aspect, the technical scheme provided by the embodiment of the invention is that the intelligent charging method of the electric vehicle based on reasoning comprises the following steps: S1, collecting electric power parameters and environment parameters of an electric vehicle during charging in real time, and preprocessing to obtain original data; S2, constructing a light multi-task learning network, dividing original data into corresponding data prediction tasks, and predicting a charging scene; And S3, optimizing the prediction result based on the multi-task loss optimization mechanism to obtain a target prediction result, and outputting an optimal charging strategy of the electric vehicle based on the target prediction result and combining with the dynamic regulation strategy. According to the method, relevant parameters of the electric vehicle during charging can be collected in real time, multidimensional features in a charging scene can be comprehensively captured, a solid data support is provided for accurately predicting the charging scene, a light-weight multitasking network is built, original data are synchronously mapped into three task results of power grid state prediction, voltage trend prediction and power gear recommendation, a plurality of independent models are not required to be deployed, efficient reasoning can be achieved on embedded equipment, light-weight and multitasking parallel processing capability is considered, through a multitasking loss optimization mechanism, power grid state recognition, voltage trend prediction and gear recommendation accuracy can be cooperatively impro