CN116691418B

CN116691418B - Charging method capable of automatically distributing control power

Info

Publication number: CN116691418B
Application number: CN202310986402.2A
Authority: CN
Inventors: 李剑锋; 朱斌; 张旭; 周震霄; 李加军; 陈莉; 赵锋; 钱天翔
Original assignee: Jiangsu Chongdong Technology Co ltd
Current assignee: Jiangsu Chongdong Technology Co ltd
Priority date: 2023-08-08
Filing date: 2023-08-08
Publication date: 2023-10-24
Anticipated expiration: 2043-08-08
Also published as: CN116691418A

Abstract

The invention relates to the technical field of intelligent power distribution, in particular to a charging method capable of automatically distributing control power. The method comprises the following steps: collecting automobile power supply data through an automatic charging pile system; data preprocessing is carried out on the automobile power supply data to generate automobile power supply noise data, wherein the automobile power supply noise data comprises charging gun state data and automobile power supply demand data; carrying out semantic information conversion processing on the state data of the charging gun based on a natural language technology to generate semantic data of the state of the charging gun; feature extraction processing is carried out on the state semantic data of the charging gun, and charging gun power feature data is generated; carrying out information demand coding processing on the charging gun power characteristic data to generate charging gun power demand coding data; according to the invention, the power control model is established by carrying out noise collection and semantication processing on the power supply data, so that the charging method capable of automatically distributing control power is realized.

Description

Charging method capable of automatically distributing control power

Technical Field

The invention relates to the technical field of intelligent power distribution, in particular to a charging method capable of automatically distributing control power.

Background

In the past, the traditional charging method is usually fixed power charging, dynamic adjustment cannot be performed intelligently according to the power grid load and the user demand, the method has various problems such as uncertain charging time, low charging efficiency, overhigh power grid load and the like, along with development of energy management and intelligent power grid technology, the charging method capable of automatically distributing control power is developed, an intelligent charging pile has communication capability and a data processing function, information such as the charging demand, the power grid load and the user priority can be monitored in real time, the charging pile is connected with equipment such as a power grid and a user terminal by utilizing the internet of things technology, real-time data transmission and information exchange are realized, the charging demand of the user and the power grid load condition are predicted by analyzing historical data and real-time data and adopting methods such as statistics and machine learning, the power output of the charging pile is automatically distributed according to a prediction result and a set priority rule, and the charging speed is flexibly adjusted, so that the user demand is met to the maximum and the stable operation of the power grid is ensured. However, the prior art still has some defects, and the charging demand prediction model still has room for improvement, especially in consideration of the complexity of user behavior and the processing and variation of mass data, and the prediction accuracy of the model needs to be improved.

Disclosure of Invention

Based on this, it is necessary to provide a charging method capable of automatically distributing control power to solve at least one of the above technical problems.

To achieve the above object, a charging method capable of automatically distributing control power, the method comprising the steps of:

step S1: collecting automobile power supply data through an automatic charging pile system; data preprocessing is carried out on the automobile power supply data to generate automobile power supply noise data, wherein the automobile power supply noise data comprises charging gun state data and automobile power supply demand data;

step S2: carrying out semantic information conversion processing on the state data of the charging gun based on a natural language technology to generate semantic data of the state of the charging gun; feature extraction processing is carried out on the state semantic data of the charging gun, and charging gun power feature data is generated; carrying out information demand coding processing on the charging gun power characteristic data to generate charging gun power demand coding data;

step S3: carrying out power demand prediction processing on the charging gun power demand coding data according to a time sequence analysis technology to generate charging gun power prediction data; deep learning processing is carried out on the charging gun power prediction data based on the long-term and short-term memory network, so that charging gun real-time power prediction data are generated;

Step S4: performing model construction processing on the automobile power supply demand data by using a Q-learning reinforcement learning method to generate a self-adaptive charging control strategy model; the real-time power prediction data of the charging gun is imported into the self-adaptive charging control strategy model through a dynamic characteristic weight adjustment formula to be dynamically adjusted, and a charging power control strategy model is generated;

step S5: performing model optimization processing on the charging power control strategy model based on a meta learning technology to generate a charging power control strategy optimization model; performing strategy optimization processing on the charging power control strategy optimization model according to the historical performance guiding optimization formula to generate charging power element learning optimization data;

step S6: performing unsupervised training treatment on the charging power control strategy optimization model according to the charging power element learning optimization data through unsupervised learning to generate a charging power control strategy supervision model; performing result adjustment processing on the charge power control strategy supervision model by using a model interpretation technology to generate a charge control strategy;

step S7: performing power distribution processing on the charging gun according to a charging control strategy to generate charging gun power control data; performing implementation result collection processing on the charging gun power control data based on a real-time data stream processing technology to obtain charging gun control implementation result data; and carrying out system circulation feedback control processing on the charging gun control implementation result data according to a feedback control theory, so as to realize automatic distribution of charging gun power control.

According to the invention, the automatic charging pile system is used for collecting the automobile power supply data, and the data preprocessing is carried out on the automobile power supply data, so that noise and abnormal values in the data can be removed, the accuracy and reliability of the data are improved, the automobile power supply noise data are generated, the noise interference in an actual charging environment can be simulated, the robustness and performance of a charging power optimization algorithm can be evaluated, the automatic charging pile system is used for collecting the data, the automatic collection and processing can be realized, and the efficiency and accuracy of the data collection and processing are improved; the method has the advantages that the semantic information conversion processing is carried out on the state data of the charging gun through the natural language technology, the data can be converted into interpretable semantic information, the readability and the understandability of the data are improved, the characteristics related to the charging power can be extracted through the characteristic extraction processing on the state semantic data of the charging gun, useful characteristic data are provided for the optimization of the subsequent charging power, the data can be encoded into a standardized data format through the information demand encoding processing on the power characteristic data of the charging gun, the processing and the analysis of a subsequent algorithm are convenient, the problems of high data dimension, difficult data understanding and the like can be overcome, and the efficiency and the accuracy of the data analysis are improved; the power demand prediction processing is carried out on the charging gun power demand coding data through a time sequence analysis technology, so that the charging power demand in a future period of time can be predicted, useful reference information is provided for charging power optimization, the deep learning processing is carried out on the charging gun power prediction data based on a long-short-period memory network, the accuracy and precision of charging power prediction can be improved, a more accurate prediction result is provided for charging power optimization, the future charging power trend can be analyzed according to historical data and the current state through generating the charging gun real-time power prediction data, a decision maker is helped to formulate a more reasonable charging strategy, the management and control of charging power can be helped to be optimized, the utilization efficiency of a charging pile is improved, the charging cost is reduced, and the user experience is improved; the Q-learning reinforcement learning method is utilized to carry out model construction processing on automobile power supply demand data, a self-adaptive charging control strategy model can be constructed according to historical data and current states, useful reference information is provided for charging power optimization, real-time power prediction data of a charging gun are imported into the self-adaptive charging control strategy model through a dynamic characteristic weight adjustment formula to carry out dynamic adjustment, the charging control strategy can be dynamically adjusted according to a real-time charging power demand prediction result, accuracy and precision of charging power control are improved, after the charging power control strategy model is generated, self-adaptive control can be carried out on charging power according to the model, charging efficiency is optimized, utilization efficiency of a charging pile is improved, charging cost is reduced, and user experience is improved; the charging power control strategy model is subjected to model optimization processing based on a meta-learning technology, so that the effect and performance of a charging power control strategy can be improved, management and control of charging power are further optimized, the charging power control strategy optimization model is subjected to strategy optimization processing according to a historical performance guidance optimization formula, the performance and accuracy of the charging power control strategy can be improved according to historical data and experience, and after charging power meta-learning optimization data are generated, the charging power control strategy can be further optimized according to the data, and the charging efficiency and user experience are improved; the method has the advantages that the effect and performance of the charging power control strategy optimization model can be further improved through the unsupervised training treatment of the charging power control strategy optimization model according to the charging power element learning optimization data, a more accurate control strategy is provided for the charging power optimization, the result adjustment treatment is carried out on the charging power control strategy supervision model by using a model interpretation technology, a decision maker can understand the prediction result and the control strategy of the model in a visual and interpretation model mode, the reliability and the reliability of decision are improved, the charging power can be dynamically controlled according to the strategy after the charging control strategy is generated, and the charging efficiency and the user experience are further improved; after the power distribution processing is carried out on the charging gun according to the charging control strategy, the power output of the charging gun can be adjusted in real time according to the charging power requirement, the charging efficiency and the user experience are improved, the result collection processing is carried out on the charging gun power control data based on the real-time data flow processing technology, the result data of the charging gun power control implementation can be monitored and collected in real time, after the system circulation feedback control processing is carried out on the charging gun control implementation result data according to the feedback control theory, the charging gun power output can be dynamically adjusted according to the real-time feedback information, the charging efficiency and the user experience are further improved, the automatic distribution and the control of the charging power are realized, the utilization efficiency of the charging pile is improved, the charging cost is reduced, and the user experience is improved. Therefore, the invention uses noise to enhance the accuracy of the power control data by regarding the noise data as valuable information and performing semantic processing, and uses a self-adaptive classification model and a meta-learning method to dynamically optimize according to the control result of the previous model, thereby further improving the power control accuracy of the model.

The invention has the advantages that the automatic charging pile system collects the automobile power supply data and preprocesses the automobile power supply data to generate the automobile power supply noise data, which is helpful for accurately predicting the power supply demand and the charging gun state, thereby improving the accuracy of charging power control, the natural language technology is used for carrying out semantic information conversion processing on the charging gun state data, extracting the charging gun power characteristic data therefrom, helping to quickly and accurately determine the charging demand, thereby improving the accuracy of charging power control, the time sequence analysis technology and the deep learning processing are used for processing the charging gun power prediction data, thereby generating the charging gun real-time power prediction data, helping to predict the future charging demand, thereby improving the accuracy of charging power control, the Q-learning reinforcement learning method is used for carrying out model construction processing on the automobile power supply demand data, generating an adaptive charge control strategy model, importing charge gun real-time power prediction data into the model for dynamic adjustment, generating a charge power control strategy model which is beneficial to dynamically adjusting charge power, further improving charge efficiency and user experience, optimizing the charge power control strategy model by using a meta learning technology, generating a charge power control strategy optimization model, optimizing the strategy according to a historical performance guiding optimization formula, generating charge power meta learning optimization data which is beneficial to improving accuracy and efficiency of the charge power control strategy, performing non-supervision training treatment on the charge power control strategy optimization model according to the charge power meta learning optimization data through non-supervision learning, generating a charge power control strategy supervision model, performing result adjustment treatment on the result adjustment model by using a model interpretation technology, generating a charging control strategy, which is beneficial to generating a more accurate charging control strategy, further improving charging efficiency and user experience, carrying out power distribution processing on a charging gun according to the charging control strategy, generating charging gun power control data, carrying out result collection processing on the charging gun power control data based on a real-time data stream processing technology, obtaining charging gun control implementation result data, carrying out system circulation feedback control processing according to a feedback control theory, realizing automatic distribution of charging gun power control, and being beneficial to automatically adjusting charging gun power output, and improving charging efficiency and user experience. Therefore, the invention uses noise to enhance the accuracy of the power control data by regarding the noise data as valuable information and performing semantic processing, and uses a self-adaptive classification model and a meta-learning method to dynamically optimize according to the control result of the previous model, thereby further improving the power control accuracy of the model.

Drawings

FIG. 1 is a flow chart illustrating a charging method capable of automatically distributing control power;

FIG. 2 is a flowchart illustrating the detailed implementation of step S2 in FIG. 1;

FIG. 3 is a flowchart illustrating the detailed implementation of step S3 in FIG. 1;

FIG. 4 is a flowchart illustrating the detailed implementation of step S4 in FIG. 1;

the achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

The following is a clear and complete description of the technical method of the present patent in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

To achieve the above object, referring to fig. 1 to 4, a charging method capable of automatically distributing control power, the method comprises the following steps:

In the embodiment of the present invention, as described with reference to fig. 1, a flow chart of steps of a charging method capable of automatically distributing control power according to the present invention is shown, and in this example, the charging method capable of automatically distributing control power includes the following steps:

in the embodiment of the invention, the automobile power supply data is transmitted from the charging gun to the charging pile system by using the proper sensor and the acquisition equipment, the automatic charging pile system acquires the automobile power supply data through the charging gun connected with the automobile, the data can comprise information such as voltage, current, frequency and the like, noise and abnormal values of the acquired automobile power supply data possibly exist, and preprocessing is needed to improve the quality and accuracy of the data, wherein the preprocessing comprises the steps of data cleaning, data integration, data regulation and the like, and the automobile power supply noise data is generated, and comprises the state data of the charging gun and the automobile power supply demand data.

in the embodiment of the invention, the state data of the charging gun can be converted into a form with more semantic information by using a natural language technology. This may involve identifying key information elements in the state of the charge gun using semantic parsing, named entity recognition, relational extraction, etc., and converting them into semantic representations, generating charge gun state semantic data, e.g. "connected" as "charge gun successfully connected to the car," unconnected "as" charge gun unconnected to the car, "extracting key feature information from the charge gun state semantic data, e.g. identifying keywords related to the charge gun state and specific operations, such as" connected "," disconnected "," charging "etc., generating charge gun power feature data, discretizing consecutive values in the charge gun power feature data, converting into codes capable of representing different levels or categories, unithermally encoding discrete classification variables in the charge gun power feature data, generating corresponding binary encoded vectors, and for charge gun state data with timing features, generating charge gun power demand encoding data using sequential encoding methods, such as cyclic neural network (RNN) or Long and Short Term Memory (LSTM) to capture timing patterns of state changes.

in the embodiment of the invention, the time sequence analysis method (such as ARIMA, exponential smoothing and the like) is used for predicting the power demand coding data of the charging gun, trends, periods and seasonal characteristics in the data are captured, prediction is carried out based on historical data, before prediction is carried out, characteristics related to the power demand can be extracted by using the characteristic engineering method, including hysteresis characteristics, statistical characteristics (such as mean value and standard deviation) and other coding characteristics of influencing factors (such as time and weather) of the historical power demand data, charging gun power prediction data is generated, and the charging gun power prediction data is input as a time sequence while other characteristics related to the charging gun power prediction data are considered. Dividing a data set into a training set, a verification set and a test set, carrying out proper data standardization processing, constructing a model by using an LSTM network, capturing long-term dependence in sequence data, adjusting by setting different LSTM layers and unit numbers according to the complexity of a problem, training the LSTM model by using the training set, updating the weight of the model by using a back propagation algorithm, for example, measuring the difference between a predicted result and an actual value by using a proper loss function (such as mean square error), evaluating the performance of the model by using the verification set, carrying out model optimization according to the evaluation result, such as adjusting a model framework, a learning rate and a regularization method, predicting the test set by using the trained and optimized LSTM model, and generating real-time power predicted data of the charging gun.

in the embodiment of the invention, the model construction process is performed on the power demand data of the automobile by utilizing a Q-learning reinforcement learning method, Q-learning is a reinforcement learning algorithm, which is used for solving the problem based on a markov decision process, an adaptive charge control strategy model is constructed by using the Q-learning method, the model learns how to balance between different states and actions so as to maximize a cumulative reward, a series of states representing the power demand is defined by encoding the power demand data of the automobile, the characteristics of the states representing the power demand can be included, the current battery capacity, the charging speed, the charging gun power demand and the like can be defined, a set of possible charge power control strategies can be defined as the action space of the model, the increase or decrease of the charging power can be included, the current power is kept unchanged and the like, the advantages and disadvantages of each state and action combination are evaluated according to factors such as the battery charging efficiency, the charging speed and the demand satisfaction degree in the charging process, the high efficiency and the demand in the charging process are satisfied by defining a reward function, and the Q-table is updated by utilizing the Q-learning algorithm, and the cumulative reward (expected value of each state and each state). And gradually optimizing the strategy model through training iteration, and introducing real-time power prediction data of the charging gun into the self-adaptive charging control strategy model through a dynamic characteristic weight adjustment formula to dynamically adjust the power prediction data so as to generate the charging power control strategy model.

in the embodiment of the invention, the model optimization processing is performed on the charging power control strategy model based on a meta-learning technology, meta-learning is a meta-level learning method of machine learning, the effect and generalization capability of a learning algorithm are improved by learning how to perform learning, the charging power control strategy model generated in the step S4 is input into the meta-learning process as an original model, a data set is divided into a training set and a test set by determining an appropriate learning task, such as historical performance index optimization, charging efficiency maximization, demand response speed and the like, so as to be used in the meta-learning optimization process, the charging power control strategy model is optimized by using the training set by using the meta-learning algorithm, the meta-learning algorithm learns how to adjust and update parameters of the original model according to the input original model and the designated learning task, so as to improve model performance, the charging power control strategy optimization model is generated, and the charging power control strategy optimization model is subjected to strategy optimization processing according to a historical performance guiding optimization formula, so as to generate charging power meta-learning optimization data.

in the embodiment of the invention, the unsupervised learning is a machine learning method, which learns the mode and structure of data from the data without labels by unsupervised learning according to the charge power element learning optimization data, processes the charge power element learning optimization data by using an unsupervised learning technology to train the charge power control strategy optimization model, uses the charge power element learning optimization data as input data, the data comprises optimization parameters and corresponding performance indexes, can be used for training the model, processes the input data by applying an unsupervised learning algorithm (such as clustering, dimension reduction and the like), learns the mode and structure of the charge power control strategy, generates a charge power control strategy supervision model, adjusts the result of the charge power control strategy supervision model by using a model interpretation technology, and uses a model interpretation technology to understand the decision process and result of the model.

In the embodiment of the invention, charging power is distributed to each charging gun based on a predefined rule, an optimization algorithm or a machine learning model, available total power is reasonably distributed to each charging gun according to factors such as charging requirements, power grid capacity and state of the charging gun, charging gun power control data are generated, a real-time data stream processing technology is used for processing the charging gun power control data, the real-time data of the charging gun are received and analyzed, a power distribution result is calculated and is processed and updated in a real-time environment, data such as actual application charging power, charging duration and electric quantity and other relevant information in the charging process are collected for each charging gun to form charging gun control implementation result data, the charging gun control implementation result data can be realized based on the charging gun control implementation result data by comparing with target power and adopting corresponding control strategies according to deviation, the power distribution of the charging gun is automatically adjusted by using a feedback control theory, the real-time adjustment of the charging gun power distribution is performed by using a feedback control mechanism in the continuous monitoring of the actual state and result in the charging process, so that the charging process is stable and efficient, and the charging requirements and power grid conditions are met.

Preferably, step S1 comprises the steps of:

step S11: collecting automobile power supply data through the automatic charging pile system;

step S12: performing missing value detection processing on the automobile power supply data, and if the automobile power supply data has missing values, performing missing value filling processing on the automobile power supply data to generate automobile power supply missing value data; performing abnormality detection processing on the automobile power supply missing value data, and if the automobile power supply missing value data has an abnormal value, performing smoothing processing on the automobile power supply missing value data to generate automobile power supply abnormality processing data;

step S13: performing data cleaning processing on the abnormal processing data of the automobile power supply to generate cleaning data of the automobile power supply;

step S14: performing data integration processing on the service cleaning data based on the data lake to generate automobile power integration data;

step S15: performing data protocol processing on the automobile power supply integrated data by using a dimension protocol method so as to generate automobile power supply protocol data;

step S16: and carrying out data noise collection processing on the automobile power supply specification data through the noise sensor so as to obtain automobile power supply noise data, wherein the automobile power supply noise data comprises charging gun state data and automobile power supply demand data.

The invention can acquire the automobile power supply data in real time through the automatic charging pile system setting, can acquire the key information of the power supply state, the battery electric quantity, the charging time length and the like of the electric vehicle, provides an accurate data basis for the subsequent charging control, detects the missing value of the acquired automobile power supply data, performs filling processing, is beneficial to ensuring the integrity and the continuity of the data, avoids the inaccuracy of the data caused by the missing value, ensures the feasibility of the subsequent data processing and analysis, performs anomaly detection, identification and potential anomaly value processing on the automobile power supply data, can reduce the influence of the anomaly data on the subsequent processing and analysis, improves the quality and the reliability of the data, performs data cleaning on the automobile power supply data after the anomaly processing, and removes repeated, invalid and wrong data, the method has the advantages that data are cleaner and more consistent, noise and interference of the data are reduced, the cleaned business data are integrated based on a data lake method, the data from different sources are integrated, comprehensive utilization of information of a plurality of data sources is facilitated, more comprehensive characteristics and indexes are extracted, a subsequent charging control decision is supported, the integrated automobile power supply data are processed by a dimension reduction method, the dimension and complexity of the data are reduced, the process of data analysis and model construction is simplified, the calculation efficiency is improved, the requirement of storage space is reduced, noise in the automobile power supply data, particularly noise of charging gun state data and automobile power supply requirement data, is learned and analyzed, and accordingly the influence of the noise on charging control and management is improved.

In the embodiment of the invention, an automatic charging pile system is built, which comprises charging pile equipment, a data acquisition module and the like, the data acquisition module is arranged to ensure that automobile power data such as battery electric quantity, charging state and the like can be accurately acquired, the acquired automobile power data is subjected to missing value detection, whether missing values exist or not is determined, if the missing values exist, missing value filling is performed by using an interpolation method or a model filling method, the integrity of the data is ensured, the filled data is subjected to anomaly detection, the abnormal values are identified, if the abnormal values exist, smoothing processing is performed, the influence of anomalies is eliminated, anomaly processing data is generated, the anomaly processing data is subjected to data cleaning, operations such as repeated data removal, anomaly data processing and effectiveness of check data are performed, the cleaned data is stored in a data lake through establishing a data lake structure, the information of different data sources is integrated, the association relation is established, the automobile power integrated data is generated, a proper dimension regulation method such as Principal Component Analysis (PCA) is selected, the integrated automobile power data is processed, the automobile power noise regulation data is generated, the automobile power noise regulation data is set, and the noise regulation data is used for collecting the corresponding power supply regulation data in the power supply regulation state data and the power supply regulation data.

Preferably, step S2 comprises the steps of:

step S21: performing data dimension reduction processing on the charging gun state data based on an NLP semantic analysis technology to generate charging gun state dimension reduction data;

step S22: carrying out semantic conversion processing on the state dimension reduction data of the charging gun according to the word embedding technology to generate a charging gun semantic vector; matrix combination processing is carried out on the semantic vectors of the charging gun, and state semantic data of the charging gun are generated;

step S23: feature extraction processing is carried out on the state semantic data of the charging gun to obtain power data of the charging gun;

step S24: performing feature analysis processing on the charging gun power data according to a variance statistical analysis method to generate charging gun power feature data;

step S25: and carrying out information demand coding processing on the charging gun power characteristic data by utilizing a binary coding technology to generate charging gun power demand coding data.

According to the invention, the data dimension reduction processing is carried out on the state data of the charging gun through the NLP semantic analysis technology, the complexity and redundancy of the data can be reduced, the data processing efficiency is improved, the state data of the charging gun can be converted into simpler and representative dimension reduction data through the NLP semantic analysis technology, important semantic information is reserved, the dimension reduction data of the state of the charging gun can be converted into semantic vectors through the word embedding technology, the semantic expression capacity of the semantic vectors is enriched, the semantic vectors of the charging gun are subjected to matrix combination processing, the information of different semantic dimensions can be combined to provide more comprehensive state semantic data of the charging gun, the characteristic extraction processing is carried out on the state semantic data of the charging gun, the power data of the charging gun is an important index for the performance and the demand of the charging gun, the power condition of the charging gun can be better understood and analyzed through the extraction of the characteristic, the change degree and the stability of the power data of the charging gun can be helped by the feature analysis processing method, the charging gun power data can be subjected to the feature analysis, the charging gun power data can be obtained to the feature analysis, the charging gun power characteristic can be described and the statistical characteristic data can be used as the basis, the subsequent power analysis data can be provided with the binary power analysis data, the demand data can be coded according to the demand data, the demand data can be further coded, and the demand data can be coded by the data is coded according to the binary power analysis data.

As an example of the present invention, referring to fig. 2, the step S2 in this example includes:

in the embodiment of the invention, through collecting a sample data set containing state data of the charging gun, the data can comprise various state information of the charging gun, such as connection state, voltage, current and the like of the charging gun, the state data of the charging gun is converted into a text representation form, for example, different state information can be combined into a descriptive text, or each state information is used as independent text data, the state data of the charging gun is subjected to semantic analysis by using an NLP technology, such as part-of-speech labeling, named entity recognition, syntax analysis and the like, so as to extract semantic information in the data, and the extracted features are subjected to dimension reduction processing by using a principal component analysis or linear discriminant analysis algorithm so as to reduce the dimension of feature vectors, so that the state dimension reduction data of the charging gun is generated.

In the embodiment of the invention, by selecting a proper Word embedding technology, such as Word2Vec, gloVe or FastText, the technologies can map words into a continuous vector space, so that words with similar semantics are closer in distance in the vector space, the selected Word embedding technology is used for carrying out semantic conversion on the state-reduced data of the charging gun to generate the semantic vector of the charging gun, each state-reduced data is used as a text input, the state-reduced data is converted into a vector representation with fixed dimension through the Word embedding technology, and the generated semantic vector of the charging gun is subjected to matrix combination processing to generate the semantic data of the charging gun, which can form a matrix data structure by combining the vectors in columns or rows.

in the embodiment of the invention, by performing feature extraction processing according to the selected features, various machine learning and signal processing technologies, such as methods of statistical feature extraction, frequency domain analysis, time domain analysis, wavelet transformation and the like, are used, the feature extraction aims to extract information and representative features from data, the power data of the charging gun is obtained by calculating according to the features subjected to the feature extraction processing, and the power value can be defined according to actual requirements, such as by calculating, aggregating or converting the features.

in the embodiment of the invention, the charging gun power data processed in the step S23 are prepared to ensure that the data format is correct and available, the data can comprise power values of the charging gun at different time points, the charging gun power data is subjected to feature analysis processing according to a variance statistical analysis method, the variance analysis is a statistical method used for checking whether the average values among different groups have significant differences, the variance analysis can be used for analyzing the features in the charging gun power data, the charging gun power data is set in groups according to actual requirements, for example, the data can be grouped according to time periods, geographic positions, charging pile types and the like, and the variance is calculated for the power data in each group. The variance represents the discrete degree of the data, the larger variance means the higher degree of the data scattering in the group, the variance of the power data between different groups is calculated, so that the degree of the difference of the power data between different groups can be compared, the variance analysis is performed according to the calculated intra-group variance and inter-group variance, whether the power characteristics between different groups have significant differences can be determined through the variance analysis, the characteristic extraction and generation can be performed on the power data of the charging gun according to the result of the variance analysis, and the power characteristic data of the charging gun can be generated, for example, the group or the characteristic with the significant differences can be selected for further analysis and processing.

In the embodiment of the invention, continuous power characteristic values can be converted into discrete value sets by discretizing the power characteristic data of the charging gun, binary coding is convenient, the number of coded bits is determined according to the value range of the discretized power characteristic data, the number of bits is enough to represent all discrete value values, the discretized power characteristic data is coded by using a binary coding technology, for example, each discrete value can be mapped into a corresponding binary code by using a standard binary coding method, the power demand coding data of the charging gun is generated according to the binary coding result, and each discretized power characteristic value is replaced by the corresponding binary code.

Preferably, step S3 comprises the steps of:

step S31: carrying out time sequence cutting processing on the charging gun power demand coded data by a sliding window method to generate time sequence cutting data;

step S32: carrying out seasonal trend decomposition processing on the time series cutting data by using a seasonal decomposition time series method to generate time series decomposition data;

Step S33: performing time sequence prediction processing on the time sequence decomposition data based on the autoregressive moving average model to generate charging gun power prediction data;

step S34: carrying out data standardization processing on the charging gun power prediction data according to the deep learning frame to generate standard charging gun power prediction data;

step S35: model training processing is carried out on a preset sequence model based on a long-short-term memory network through standard charging gun power prediction data, and a charging time sequence prediction model is generated; and importing the standard charging gun power prediction data into a charging time sequence prediction model to perform power prediction processing, and generating charging gun real-time power prediction data.

The invention cuts the time sequence by a sliding window method, then applies a seasonal decomposition method, can help to decompose the time sequence data into three parts of trend, seasonal and residual, and the decomposition is helpful to understand the structure and mode of the time sequence data, provides a basis for subsequent prediction and modeling, and can use an autoregressive moving average model (ARIMA) or other time sequence prediction methods to predict the power of the charging gun by using the time sequence decomposition data. The method and the system enable future power demands to be presumed according to historical data and trends, are favorable for reasonable scheduling and resource management of the charging equipment, can enable the data to have similar scales and ranges through standardized processing of charging gun power prediction data, avoid deviation caused by differences among different features, are beneficial to improving training effects and performances of models, enable the models to more accurately conduct charging power prediction, utilize the standardized power prediction data to train a preset long-short-term memory (LSTM) model, can establish a charging time sequence prediction model, effectively capture long-term dependence of the data, input the standardized power prediction data into the trained charging time sequence prediction model, can generate innovative charging gun power prediction data, can help a decision maker to better know change trend of the future charging demands, and accordingly take corresponding measures, such as adjusting deployment of the charging equipment and optimizing energy scheduling.

As an example of the present invention, referring to fig. 3, the step S3 in this example includes:

in the embodiment of the invention, by preparing a time series data set containing the charging gun power requirement coding data, ensuring that the data are arranged according to time sequence, selecting a proper sliding window size, wherein the window is used for cutting the time series data, the window size is selected according to specific application scenes and data characteristics, starting from the starting point of the time series data, sliding with a window with fixed size, scanning the whole time series data set in sequence, each sliding window moves by one step, extracting a subsequence of a window from the data set as time series cutting data, wherein the subsequences can be continuous time windows or windows divided according to specific rules, and repeating the sliding window movement until the sliding window covers the whole time series data set, so as to generate all the time series cutting data.

In the embodiment of the invention, the seasonal period is determined by observing the data set, and the period can be daily, weekly, monthly or other specific time intervals, and is selected according to the characteristics of the data, and a seasonal decomposition method, such as an addition model or a multiplication model, is used for decomposing the time sequence cutting data. Common decomposition methods include seasonal decomposition, moving average decomposition, etc., which decompose a time series into trend, seasonal and random components, extract trend components from the decomposition result, the components representing long-term trend changes in the data, the trend components reflecting patterns of changes in the data on a longer time scale, extract seasonal components from the decomposition result, the components representing periodic fluctuations in the data, the seasonal components reflecting repeated patterns of the data over the seasonal period, extract random components from the decomposition result, generate time series decomposition data, the data representing random noise or unpredictable variations in the data.

In the embodiment of the invention, a proper ARMA model (autoregressive moving average model) is selected according to the characteristics of data and a prediction target, the ARMA model is a statistical model widely applied to time sequence prediction, the ARMA model is estimated by combining the concepts of Autoregressive (AR) and Moving Average (MA) through a least square method, maximum likelihood estimation or other statistical methods by using historical time sequence data, the estimated ARMA model is subjected to diagnostic test, the fitting degree and the reliability of the model are ensured, the common diagnostic method comprises the steps of observing an autocorrelation graph and a partial autocorrelation graph of a residual sequence, carrying out statistical test, carrying out time sequence prediction by using the estimated ARMA model, determining a time point to be predicted according to a predicted time range and frequency, carrying out prediction by using the historical data and the existing model parameters, and generating the predicted data of the charging gun power according to the predicted result of the ARMA model.

in the embodiment of the invention, the deep learning framework is used for carrying out data standardization processing on the charging gun power prediction data, the charging gun power prediction data is linearly mapped to the range of [0, 1], and the standardization formula is as follows:

x_std = (x - x_min) / (x_max - x_min)；

Where x is the original charging gun power prediction data, x_std is the normalized data, and x_min and x_max are the minimum and maximum values of the charging gun power prediction data, respectively, to generate the standard charging gun power prediction data.

In the embodiment of the invention, the proper LSTM model is selected as the charging time sequence prediction model, and relevant configuration required by model training is prepared, comprising super parameters and structures of the model, standardized charging gun power prediction data are divided into a training set and a testing set, the training set occupies most of the total data amount and is used for training and parameter adjustment of the model, the testing set is used for evaluating the performance of the model, the LSTM model is trained by using the training set, the weight and bias of the model are optimized by minimizing a loss function in the training process, so that the charging gun power time sequence can be predicted more accurately, and the performance of the trained model is evaluated by using the testing set. The prediction accuracy of the model can be measured by using various indexes (such as root mean square error, average absolute error and the like), and the standardized charging gun power prediction data is input into the model by using the trained charging time sequence prediction model to perform power prediction processing. This will generate innovative power prediction data for the charging gun, i.e. predicting the trend of power change over a period of time in the future from historical information.

Preferably, step S4 comprises the steps of:

step S41: carrying out data state expression processing on the automobile power supply demand data by using a Markov decision process to generate automobile power supply MDP expression data;

step S42: model training processing is carried out on the MDP expression data of the automobile belt energy source based on a Q-learning algorithm, and automobile power supply model training data are generated;

step S43: performing simulation test evaluation processing on the training data of the automobile power supply model to generate the data of the automobile power supply model; comparing the automobile power supply model data with a preset prediction result data range, and when the automobile power supply model data is out of the prediction result data range, performing data optimization on the automobile power supply model data to generate an adaptive control strategy model;

step S44: the real-time power prediction data of the charging gun is imported into a characteristic weight adjustment formula to perform characteristic weight calculation processing, so as to obtain charging power characteristic weight data;

step S45: and dynamically adjusting the self-adaptive charging control strategy model according to the charging power characteristic weight data, so as to generate the charging power control strategy model.

According to the invention, the data state expression processing is carried out on the automobile power supply demand data by utilizing the Markov decision process to generate the automobile power supply MDP expression data, so that the original data can be converted and expressed, the original data is suitable for modeling of an MDP model, the dynamic and state change of the automobile power supply demand can be better understood and described, and the subsequent model training and control strategy generation are facilitated; model training processing is carried out on the MDP expression data of the automobile belt energy based on a Q-learning algorithm, automobile power supply model training data are generated, a reinforcement learning algorithm is utilized to train a model, so that the model can make an optimal decision according to the current state and learn a strategy suitable for automobile power supply management, and through model training, data capable of capturing dynamic characteristics of a system and supporting the decision can be obtained, so that a foundation is laid for subsequent simulation test and evaluation; the automobile power supply model training data are subjected to simulation test evaluation processing to generate automobile power supply model data, and the trained models are subjected to simulation test, so that power supply requirements and control strategies under different situations can be simulated, performance of the models under the situations can be evaluated, and the effectiveness and feasibility of the models can be verified; the real-time power prediction data of the charging gun is imported into a characteristic weight adjustment formula to perform characteristic weight calculation processing to obtain charging power characteristic weight data, and the importance degree of each characteristic on the charging power can be known by calculating the weight of the charging power characteristic, so that guidance and basis are provided for generation of a charging power control strategy; the self-adaptive charging control strategy model is dynamically adjusted according to the charging power characteristic weight data, so that the charging power control strategy model is generated, the control strategy model can be adjusted and optimized according to the characteristic weight information, the charging power is more reasonably controlled, dynamic decision is carried out according to real-time requirements and situations, and the efficiency, stability and adaptability of the charging process are improved.

As an example of the present invention, referring to fig. 4, the step S4 includes, in this example:

in the embodiment of the invention, by determining a suitable state definition according to the requirements of the problem and the characteristics of the data set, the state is a characteristic variable describing the problem, and the state should reflect the related information of adjacent time points, for the power supply requirement problem, which may include the residual capacity of a vehicle battery, the current charging state, the residual driving mileage of the vehicle, and the like, according to the vehicle power supply requirement data, the state change of the adjacent time points is analyzed, the state transition probabilities are calculated, and these probabilities can be obtained by a statistical method, a machine learning method or domain expert knowledge, for example, the charging behavior of the vehicle under different battery capacities can be analyzed according to the historical data, and then the transition probability from one state to another is estimated, the action executable under each state is determined, for the power supply management problem, the possible actions include starting charging, stopping charging or maintaining the current state, defining a reward function to evaluate the degree of each state and action, the reward function can be designed according to specific targets and constraints, for example, the reward function can be set according to the factors such as the service life of the battery, the charging efficiency or the user degree, and the like, the reward can be obtained by a statistical method, the transition probability, the vehicle behavior and the MDP can be analyzed according to the determined state definition, the transition probability and the vehicle state expert knowledge, the action and the MDP can be well as a model, and the optimal state is set, for the optimal state algorithm is set, and a decision algorithm is set, and an iterative algorithm is made, for solving, for the optimal algorithm is made, for the decision algorithm is made, and has been made, and has an iterative algorithm.

in the embodiment of the invention, model training processing is performed on the MDP expression data of the automobile belt energy based on a Q-learning algorithm, the Q-learning algorithm represents the state and the action value through a Q-table, firstly, an empty Q-table needs to be initialized, the row of the empty Q-table represents the state, the column represents the action, each element represents the value of executing the specific action under the specific state, and two important learning parameters exist in the Q-learning algorithm: learning rate (α) and discount factor (γ). The learning rate determines how much new information will be learned each time the Q value is updated, the discount factor determines how much attention will be paid to future returns, and the Q-learning algorithm uses the epsilon-greedy strategy to select actions based on the characteristics and needs of the problem. In the early stage, exploratory strategies are adopted, namely actions are randomly selected with certain probability; and in the later stage, the Q value calculated by the Q-table is used for decision making. Setting proper exploration probability according to the requirements of the problem, and iteratively updating the Q value: the Q value is started to be iteratively updated, and the specific steps are as follows:

1. Observing the current state, and selecting actions to be executed according to the current state and the action strategy;

2. performing the selected action and observing the next status and the harvested rewards;

3. according to the updating rule of the Q-learning algorithm, updating the Q value of the corresponding state and action in the Q-table, wherein the updating formula is as follows:

Q(s, a) = (1 - α) * Q(s, a) + α * (r + γ * max(Q(s’, a’)))；

where α is the learning rate, r is the reward received after the action is performed in the current state, γ is the discount factor, and max (Q (s ', a')) represents the maximum Q value for the next state;

4. c, updating the current state to be the next state, and continuing to circulate the steps a to c until a stopping condition is reached;

and repeating iterative updating until the Q value converges or reaches a preset training round number, gradually converging the Q value in the Q-table to an optimal value in the whole training process, representing an optimal state-action pair, and recording the current state, action, rewards and the next state each time the Q value is updated in the training process, wherein the data can be used as training data of a model, namely training data of an automobile power model.

In the embodiment of the invention, the trained automobile power supply model data is used for testing in a simulation environment, the automobile power supply behaviors under different environments are simulated according to the state information and the selected control strategy in the data set, in each simulation step, the output results given by the automobile power supply model are used according to the current state and the selected control strategy, the output results can be related information such as power supply power, energy consumption and the like, model output data in the simulation process are recorded, various parameters and indexes related to the power supply behaviors are included, automobile power supply model data are generated, a preset prediction result data range is defined, the expected power supply output range, the energy consumption range and the like, the automobile power supply model data obtained through the simulation test are compared with the prediction result data range, if the data exceeds the prediction result data range, the next step of data optimization processing is carried out, the reasons for exceeding the prediction result data range can include model errors, environment changes, input data noise and the like, the appropriate data processing method is adopted for the model output data which does not meet the requirements, the data processing method can include data smoothing, abnormal value processing, abnormal value parameter processing or control strategy adjustment is carried out, and the model self-adaptive data is generated according to the control strategy and the like.

in the embodiment of the invention, the real-time power prediction data of the charging gun is imported into the characteristic weight adjustment formula by determining the formula or algorithm for calculating the characteristic weight, the characteristic weight data of the charging power is calculated according to the defined characteristic weight adjustment formula, the weight value reflects the relative contribution degree of the real-time power prediction data of the charging gun in the aspect of characteristic importance, the obtained characteristic weight data of the charging power is analyzed, the influence degree of each characteristic on the charging power is known, and the innovative power prediction model of the charging gun can be optimized or related decisions and strategies can be formulated according to the result of the characteristic weight data.

In the embodiment of the invention, the charging power characteristic weight data obtained by calculation in the step S44 are used, the data reflect the relative importance of different characteristics on the charging power, the accuracy and the reliability of the characteristic weight data are ensured, so that a reliable result is obtained when the strategy model is adjusted, and the charging power characteristic weight data are applied to the self-adaptive charging control strategy model according to the charging power characteristic weight data. Specific adjustment modes will be determined according to the design of the model, and possible adjustment modes include, but are not limited to: according to the weight value, adjusting the weight input by the feature, adjusting the parameter in the algorithm, modifying the decision rule or adjusting the weight of the machine learning model, using the verification data set or testing the adjusted charging control strategy model in the actual charging scene, analyzing the performance and effect of the model, comparing the difference before and after adjustment, ensuring that the adjusted model has better charging power control performance, and generating the charging power control strategy model.

Preferably, the characteristic weight adjustment formula in step S44 is specifically as follows:

；

in the method, in the process of the invention,expressed as a charging power characteristic weight, +.>Denoted as charging gun charging time->Denoted as a current value of the charging gun during charging, ">Transmission line voltage, denoted charging gun +.>Expressed as charging temperature of the charging pile->Expressed as car battery capacity>Expressed as the actual current value of the car battery, +.>Expressed as a charging power factor>Is expressed as ambient humidity when charging the charging gun, +.>Represented as a charging power characteristic weight abnormality adjustment value.

The characteristic weight adjustment formula is used for deriving the charging time through the flowing current value when the charging gun is charged and the ambient humidity when the charging gun is charged so as to measure the influence of the charging time change on the charging power. In practical application, the formula can calculate and adjust the weight of the charging power characteristics according to the input innovative power prediction data so as to determine the contribution degree of each characteristic to the charging power. The formula fully considers the charging time of the charging gun The value of the flowing current when charging the charging gun>Power transmission line voltage of charging gun>Charging temperature of charging pile->Automobile battery capacity->Actual current value of automobile battery +.>Charging power factorAmbient humidity during charging of charging gun>Charging power characteristic weight abnormality adjustment value +.>According to the current value +.>The interrelationship between the above parameters constitutes a functional relationship:

；

the characteristic weight can be adjusted by knowing the limit relation between the battery capacity and the battery current through the interaction relation between the battery capacity and the actual current value of the automobile battery, the characteristic weight is adjusted under the condition of ensuring the accuracy of regional data, the data redundancy is reduced under the condition of ensuring the accuracy of the data by utilizing the charging power factor,saving calculation force, enabling calculation to achieve rapid convergence, and adjusting value by abnormal weight of charging power characteristicsThe generation of the charging power characteristic weight result is adjusted, and the charging power characteristic weight is generated more accurately>The accuracy and the reliability of characteristic weight adjustment are improved. Meanwhile, parameters such as a charging power factor, the ambient humidity when the charging gun is charged and the like in the formula can be adjusted according to actual conditions, so that different characteristic weight adjustment scenes are adapted, and the applicability and the flexibility of the algorithm are improved.

Preferably, step S5 comprises the steps of:

step S51: performing meta-learning processing on the charging power control strategy model according to model engine driving learning to generate a charging power meta-learning capture environment;

step S52: the charging power element learning and capturing environment is used for autonomously capturing charging gun power prediction data, and when the charging power element learning and capturing environment detects that the charging gun power prediction data is smaller than the preset standard power, the charging power is distributed; when the charging power element learning capture environment detects that the charging gun power prediction data is smaller than the preset standard power, priority ranking is carried out on the automobile power supply demand data, and high-priority automobile power supply demand data and low-priority automobile power supply data are generated;

step S53: when the charging power control strategy model detects high-priority automobile power supply demand data and low-priority automobile power supply data, preferentially supplying power to an automobile corresponding to the high-priority automobile power supply demand data, so as to generate a charging power control strategy optimization model;

step S54: and carrying out strategy optimization processing on the charging power control strategy optimization model by utilizing a historical performance guiding optimization formula to generate charging power element learning optimization data.

According to the invention, the charging power control strategy model is subjected to meta learning processing according to model engine driving learning, so that the model can learn from experience, information and characteristics in various environments are integrated, and the adaptability and generalization capability of the model in a new environment are improved; by autonomously capturing and processing the charging gun power prediction data, the system can dynamically allocate the charging power according to actual conditions, balance and efficiency of a charging process are ensured, priority ordering processing can ensure that the power supply requirement of a high-priority automobile is met, and user experience and reliability of charging service are enhanced; by supplying power to the high-priority automobiles with priority, the system can meet the charging requirements of important users or key requirements, the availability and user satisfaction of the system are improved, and the charging power control strategy optimization model can adapt to different requirements and environmental conditions through dynamic adjustment, so that flexible and intelligent control is realized; by utilizing the historical performance guiding optimization formula, the charging power control strategy can be effectively optimized and improved, the efficiency, stability and reliability of the charging process can be improved, and the charging time and energy waste can be reduced.

In the embodiment of the invention, a training data set is prepared by determining a meta-learning algorithm and a model engine, such as a neural network model in deep learning, comprising charging power data and environmental characteristic data, the training data set is used for meta-learning training on the model, parameters and weights of the model are adjusted, so that the model can learn and capture characteristics and rules of the environment from the data, the performance of the trained meta-learning model is verified and evaluated, the model can accurately capture the characteristics of the charging environment, a charging power meta-learning capture environment is generated, a sensor and monitoring equipment in a charging system are configured to acquire charging gun power prediction data, the charging gun power prediction data is monitored and collected in real time based on the meta-learning model capture environment, whether the charging gun power prediction data is smaller than preset standard power or not is judged, if the charging power prediction data is smaller than the standard power, the charging power is dynamically distributed and adjusted according to strategies, the stability and efficiency of a charging process are ensured, meanwhile, the vehicle power demand data is subjected to priority ranking treatment, the vehicle power demand data is divided into high priority and low priority vehicle power demand data, whether the vehicle power demand data is high priority exists or not is judged according to the output result of the charging power control strategy model, whether the high priority vehicle power demand data exists or low priority data is met, the charging power demand data is optimized, the charging power demand data is calculated, the charging demand performance is optimized, the charging power demand data is calculated, and the charging performance is optimized, and the charging power demand data is calculated, and the charging performance is based on the priority data is optimized, and the charging power demand data is based on the priority and the performance has been optimized, and the charging performance data has a priority, the method is used for guiding the optimization processing of the charging power control strategy model, carrying out strategy optimization on the charging power control strategy optimization model by using a historical performance guiding optimization formula, adjusting parameters and weights of the model, generating charging power element learning optimization data through the optimization processing, and improving and optimizing the training and deducing process of the charging power control strategy model.

Preferably, the historical performance guidance optimization formula in step S54 is specifically as follows:

；

in the method, in the process of the invention,expressed as optimized charging power, +.>Reference constant expressed as adjustment final optimization result, < ->An angle value expressed as a guaranteed denominator +.>Expressed as the number of historical performance data, +.>Expressed as number of time steps>Expressed as actual charging power +.>Charging power, denoted as last time step, ">Expressed as +.>The average value of the individual data is calculated,expressed as +.>Standard deviation of individual data>Expressed as predicted charging power +.>The anomaly correction is optimized as a historical performance guide.

The invention constructs a historical performance guidance optimization formula for the number of the historical performance data and the first historical performance dataThe standard deviation of the data is used for smoothing the historical performance data to measure the average value of the historical performance data, the historical performance guiding optimization formula can be used for deriving a periodical weighted average term according to the number of time steps and the charging power of the last time step, representing the fluctuation of the actual charging power in time and according to the actual charging power and the +. >The average of the data is used for evaluating the average deviation degree of the actual charging power in time, so that the optimized charging power is accurately determined. In practice, the formula may be in the form of historical performance dataOn the basis of (1) solving an optimal charging power control strategy by using an optimization algorithm, wherein a denominator part is an average value of historical performance data, a numerator part is an average value of predicted charging power, and an angle value +.>The method is used for guaranteeing that the value of the denominator is not zero, and avoiding the situation of dividing by zero. Constant->For adjusting the size of the final optimization result. The formula fully considers the reference constant for adjusting the final optimization result +.>Ensure the angle value of denominator +.>Number of historical performance data ∈>Time step->Actual charging power +.>Charging power of last time stepThe historical performance data +.>Mean>The historical performance data +.>Standard deviation>Predicted charging power +.>Historical performance-guided optimization of abnormal correction amount +.>According to the number of historical performance data +.>The interrelationship between the above parameters constitutes a functional relationship:

；

the magnitude of the adjustment and optimization result can be known by adjusting the interaction relation between the reference constant of the final optimization result and the angle value of the guaranteed denominator, the historical performance guidance optimization is performed under the condition of ensuring the accuracy of the regional data, the predicted charging power is utilized, the data redundancy is reduced under the condition of ensuring the accuracy of the data, the calculation force is saved, the calculation is enabled to achieve rapid convergence, and the abnormal correction quantity is optimized through the historical performance guidance The generation of the optimized charging power result is regulated, and the optimized charging power is generated more accurately>The accuracy and the reliability of the historical performance guiding optimization are improved. Meanwhile, parameters such as an angle value of a guaranteed denominator, a reference constant of a final optimization result of the charging adjustment and the like in the formula can be adjusted according to actual conditions, so that different historical performance guiding optimization scenes are adapted, and applicability and flexibility of the algorithm are improved.

Preferably, step S6 comprises the steps of:

step S61: based on the self-encoder, the charging power element learning optimization data is imported into a charging power control strategy optimization model to carry out unsupervised training treatment, and a charging power control strategy supervision model is generated;

step S62: adopting a local interpretable model to perform prediction result interpretation on the charging power control strategy supervision model to generate a charging power control interpretation result;

step S63: and carrying out result adjustment processing on the charge power control strategy supervision model by using the charge power control interpretation result to generate a charge control strategy.

According to the invention, the self-encoder is used for importing the charging power element learning optimization data into the charging power control strategy optimization model for performing unsupervised training, the training method can learn hidden features and modes from the data without explicit labels or supervision signals, and as a result, a charging power control strategy supervision model is generated, and key factors and rules in the charging process can be better understood and captured by the model; the prediction result interpretation is carried out on the charge power control strategy supervision model by adopting the local interpretable model, the interpretation models can help to interpret the prediction result of the charge power control strategy model, reveal the reason and basis of specific prediction made by the model, are beneficial to increasing the understanding and the interpretability of decisions in the charging process, and improve the credibility and the acceptability of the system; the prediction result interpretation is carried out on the charge power control strategy supervision model by adopting the local interpretable model, the interpretation models can help to interpret the prediction result of the charge power control strategy model, reveal the reason and the basis of specific prediction made by the model, are beneficial to increasing the understanding and the interpretability of the decision in the charging process, and improve the credibility and the acceptability of the system.

In the embodiment of the invention, through collecting data related to a charging process, including charging current, voltage, temperature and the like, selecting a proper self-encoder structure, using the collected data for training a self-encoder model, the self-encoder aims at learning potential representation of input data and reconstructing original data as much as possible, using a trained self-encoder model, inputting potential representation obtained by charging power element learning optimization data through the encoder into a charging power control strategy optimization model for performing unsupervised training, optimizing and verifying the charging power control strategy supervision model, ensuring that the charging power control strategy supervision model can accurately learn and capture useful characteristics and rules in the charging process, generating the charging power control strategy supervision model, and selecting a local interpretable model: for example, a local interpretability model, such as LIME (local interpretation model) or SHAP (measurable feature importance), may be used to interpret the prediction results of the charge power control policy supervision model, the prediction results of the charge power control policy supervision model are input into the local interpretation model, the contribution degree of each input feature in the prediction results to the prediction results is interpreted, a charge power control interpretation result will be generated, the cause and basis of the prediction results are helped to understand, by analyzing the charge power control interpretation result, which features play an important role in the control policy in the charging process is determined, the charge power control policy supervision model is adjusted according to the interpretation result, the weight, parameters or structure of the model may be adjusted to further optimize the performance and adaptability of the model, and a final charge control policy is generated according to the adjusted charge power control policy supervision model, which is capable of dynamically adjusting the charge power according to the interpretation result to meet the requirements and objectives in the charging process.

Preferably, step S7 comprises the steps of:

step S71: setting power distribution parameters of the charging gun according to a charging control strategy, and generating power control parameters of the charging gun;

step S72: applying the charging gun power control parameters to an actual charging gun to perform power control operation, and generating a charging gun power control operation result;

step S73: acquiring and processing a charging gun power control operation result by using a real-time data stream processing technology, so as to acquire real-time charging gun power control data;

step S74: carrying out data analysis processing on the real-time charging gun power control data to generate charging gun control implementation result data; performing feedback control processing on the charging gun control implementation result data according to a feedback control theory to generate a feedback control instruction;

step S75: and applying the feedback control instruction to the charging gun to realize automatic distribution of the power of the charging gun.

According to the invention, the reasonable utilization and power distribution of the charging gun can be realized by setting the power distribution parameters of the charging gun according to the charging control strategy; the generated power control parameters of the charging gun are applied to actual charging gun equipment, the power control operation of the charging gun is realized through a corresponding control mechanism, an adjusting device or a software system, and the power output by the charging gun can be controlled according to the set power distribution parameters; monitoring and collecting power control operation of the charging gun by using a real-time data stream processing technology, and acquiring power control data of the charging gun in real time, including parameters such as actual output power, voltage, current and the like, through a sensor, a monitoring system or other data acquisition equipment; analyzing and processing the power control data of the charging gun acquired in real time, and generating charging gun control implementation result data by using data analysis, machine learning or other related technologies, wherein the data reflects the power control condition of an actual charging gun, and based on a feedback control theory, the control implementation result data is used as feedback information to regulate and optimize the power control of the charging gun to generate a feedback control instruction; the generated feedback control instruction is applied to the charging gun equipment, the power distribution of the charging gun is automatically regulated through a control mechanism, an algorithm or a software system, the power of the charging gun can be dynamically distributed according to actual charging requirements and power grid conditions, the optimal control and the efficiency improvement in the charging process are realized, the automation and the optimization of the power control of the charging gun are realized, the charging efficiency is improved, the power grid conditions are adapted, and the reasonable distribution and the stable operation of the charging process are ensured.

In the embodiment of the invention, a charging control strategy is formulated according to requirements and grid conditions, for example, the power of a charging gun is distributed according to factors such as a vehicle state and a grid load, the power distribution parameters of the charging gun are set according to the charging control strategy, the maximum power or the power range which can be distributed by each charging gun is determined, a control command is sent to charging gun equipment according to the set power distribution parameters, the power output of the charging gun is regulated, parameters such as the current, the voltage and the power of the charging gun are monitored in real time by using a monitoring system or a sensor, the charging gun is ensured to work according to the set power control parameters, the result of the charging gun power control operation is recorded, namely, the charging gun power control operation result is generated, the information such as the actual output power and the timestamp is generated, the real-time data including the power, the current and the voltage is acquired by using a real-time data stream processing technology, the acquired real-time data including the power, the power of the parameters such as the power, the current and the voltage is processed and converted to adapt to the subsequent analysis and the control requirements, the real-time charging gun power control data is analyzed and processed, for example, the deviation of the calculated power distribution, the charging load prediction and the like is performed, the charging gun power control data is correspondingly, the charging gun power control command is processed, the power control command is generated, the power control device is automatically, the power control gun is continuously is controlled and the power control command is continuously is regulated according to the feedback control command is carried out, or the feedback control command is realized, the power control command is realized, which is continuously is adjusted, or the power control device is realized, and the charging control device is continuously is adjusted according to the real-time control command is continuously or is adjusted.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A charging method capable of automatically distributing control power, comprising the steps of:

2. The charging method capable of automatically distributing control power according to claim 1, wherein the step S1 comprises the steps of:

3. The charging method capable of automatically distributing control power according to claim 2, wherein the step S2 comprises the steps of:

4. A charging method capable of automatically distributing control power according to claim 3, wherein step S3 comprises the steps of:

5. The charging method capable of automatically distributing control power according to claim 4, wherein step S4 comprises the steps of:

6. The method of claim 5, wherein the characteristic weight adjustment formula in step S44 is as follows:

；

7. The charging method capable of automatically distributing control power according to claim 5, wherein step S5 comprises the steps of:

8. The charging method capable of automatically distributing control power according to claim 7, wherein the historical performance guidance optimization formula in step S54 is as follows:

；

In the method, in the process of the invention,expressed as optimized charging power, +.>Reference constant expressed as adjustment final optimization result, < ->An angle value expressed as a guaranteed denominator +.>Expressed as the number of historical performance data, +.>Expressed as number of time steps>Expressed as actual charging power +.>Charging power, denoted as last time step, ">Expressed as +.>Mean of the individual data>Expressed as +.>Standard deviation of individual data>Expressed as predicted charging power +.>The anomaly correction is optimized as a historical performance guide.

9. The charging method capable of automatically distributing control power according to claim 7, wherein step S6 comprises the steps of:

10. The charging method capable of automatically distributing control power according to claim 9, wherein step S7 comprises the steps of: