Artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method, system, device and storage medium

Abstract

The application provides an artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method, system, device and storage medium. The method comprises regularly collecting power grid load data in a preset time period of a preset power consumption area, wherein the power grid load data comprises first load data, electric vehicle charging demand data, user behavior data and weather data; the power grid load data is subjected to data cleaning, feature extraction and data fusion to generate a fusion data set; a first algorithm model processes the fusion data set to obtain power grid load prediction data; a charging and discharging strategy is dynamically generated based on the power grid load prediction data and a DQN algorithm; and the power grid power supply is dispatched to the electric vehicle battery or the electric vehicle battery power supply is dispatched to the power grid according to the charging and discharging strategy. The method can control electric vehicle charging and discharging, balance power grid load fluctuation and improve power resource utilization rate based on load prediction and optimized scheduling model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of smart grid, and particularly relates to a method, system and device for intelligent scheduling of electric vehicle bidirectional charging power grid load based on artificial intelligence and a storage medium. BACKGROUND

[0002] With the rapid development of electric vehicles and the surge in charging demand, the traditional power grid faces greater load fluctuation pressure, especially during peak electricity consumption periods, making it difficult to effectively manage a large number of charging demands, with the power grid load being too heavy. High load leads to voltage changes, affecting the stability of household appliances and industrial equipment operation, as well as the stability of the power grid. At the same time, electric vehicles have large-scale energy storage potential. If bidirectional charging and discharging interaction with the power grid can be achieved, not only can the fluctuation burden of the power grid be reduced, but also new benefits can be brought to electric vehicle users. The existing load management method lacks intelligent prediction capability and is difficult to schedule the charging and discharging operations of electric vehicles in real time, so there is an urgent need for an innovative system that integrates artificial intelligence load prediction, bidirectional charging and discharging management, and intelligent scheduling. SUMMARY

[0003] The present application discloses a method and system for intelligent scheduling of electric vehicle bidirectional charging power grid load based on artificial intelligence, which can control electric vehicle charging and discharging based on load prediction and optimized scheduling model, balance power grid load fluctuation, and improve power resource utilization.

[0004] In a first aspect, the present application provides a method for intelligent scheduling of electric vehicle bidirectional charging power grid load based on artificial intelligence, comprising:

[0005] Periodically collecting power grid load data in a preset time period in a preset electricity consumption area, the power grid load data including first load data, electric vehicle charging demand data, user behavior data, and weather data;

[0006] Performing data cleaning, feature extraction, and data fusion on the power grid load data to generate a fusion data set;

[0007] A first algorithm model processes the fusion data set to obtain load prediction data of the power grid;

[0008] Generating a charging and discharging strategy based on the load prediction data of the power grid and a DQN algorithm;

[0009] According to the charging and discharging strategy, scheduling power supply from the power grid to the electric vehicle battery, or scheduling power supply from the electric vehicle battery to the power grid.

[0010] In some embodiments, the first load data includes voltage data, current data, power data, and electricity consumption period data.

[0011] In some embodiments, the power data includes active power data, reactive power data, power factor data.

[0012] In some embodiments, the weather data includes temperature data, humidity data, wind speed data, rainfall data, snowfall data, lightning data, sunshine intensity data, haze data, sand data.

[0013] In some embodiments, the user basic information includes user quantity data, user distribution data, user type data, user travel habit data.

[0014] The vehicle information includes vehicle population data, vehicle battery parameter data, vehicle range data.

[0015] The charging behavior information includes charging time data, charging location preference data, charging frequency data.

[0016] The market and policy data includes market growth rate data, policy guidance data, wherein the market growth rate data includes load size change data and load space-time distribution change data, and the policy guidance data includes data of user charging strategy change caused by electricity price policy.

[0017] In some embodiments, the user behavior data includes timestamp data, time period data, instantaneous power data, average power data, device classification data, device quantity data, user category data, geographic location data.

[0018] In some embodiments, the data cleaning, feature extraction, and data fusion of the power grid load data to generate the fusion data set include: supplementing missing power grid load data by mean interpolation or linear interpolation method, calculating the standard deviation and mean value of the corresponding power grid load data, regarding data points exceeding the mean value ± 2 times the standard deviation range as abnormal values, adjusting the abnormal values by analyzing the trend of adjacent normal data points, and identifying and deleting these repeated values by comparing the fields of the data records;

[0019] The data cleaning, feature extraction, and data fusion of the power grid load data to generate the fusion data set include: calculating at least one of the maximum value, minimum value, average value, and standard deviation of the power grid load data, calculating the average value of the power grid load data in a first preset time period to obtain moving average feature data, calculating the moving autocorrelation coefficient feature data packet of the first power grid load data packet at the starting time point and the second power grid load data packet at the terminal time point in a preset second time period, performing fast Fourier transform on the power grid load data to obtain Fourier transform feature values, and calculating a change rate data packet based on the power grid load data.

[0020] The data cleaning, feature extraction and data fusion of the power grid load data to generate the data fusion in the fusion data set include:

[0021] The power grid load data, the charging demand data, the user behavior data and the weather data are fused to form a fusion data set.

[0022] In some embodiments, the first algorithm model processes the fusion data set to obtain the power grid load prediction data, and the processing includes:

[0023] The fusion data set is converted into matrix data, and the matrix data is normalized and divided into a training set, a validation set and a test set.

[0024] The first algorithm model includes an LSTM model and a GRU model, the training set is input into the LSTM model and the GRU model for pre-training, the first hidden state set and the second hidden state set are obtained based on respective independent loss functions and optimizers, the fusion hidden state set is obtained by fusing the first hidden state and the second hidden state set, and the power grid load prediction data is obtained after the fusion state set is input into a shared output layer. It should be noted that the LSTM is a long short-term memory network model, the English name is Long-Short Term Memory model, and the GRU model is a gated recurrent unit, the English name is Gated Recurrent Unit.

[0025] In some embodiments, the charging and discharging strategy is dynamically generated based on the power grid load prediction data and the DQN algorithm, and the dynamically generating includes:

[0026] The current electricity price, the remaining battery capacity, the number and distribution of electric vehicles connected, the power grid carrying capacity of different regions and other information are used to construct an environment model, and the state vector S is represented as:

[0027] S=[L t ,L t+1 ,…,L t+n ,P t ,E r1 ,E r2 ,…,E rm ,N ev ,C a1 ,…,C ak ]

[0028] L t represents the current power grid load, L t+i represents the predicted load of i future time periods, wherein i is any one of 1 to n, P t represents the current electricity price, E rj represents the remaining capacity of the jth electric vehicle, wherein j is any one of 1 to m, N evrepresents the total number of electric vehicles accessing the power grid, C ah represents the power grid capacity related index of the hth region, wherein the value of h is any one of 1 to k;

[0029] define a discrete action set A, the action set A includes at least one of fast charging, slow charging, pausing charging, and battery discharging to feed back to the power grid, input the state vector S into a neural network, and output a selection probability distribution π(a|s) of each action, wherein π represents the probability distribution, a represents the action, and s represents the state;

[0030] According to the peak-valley difference of the power grid, the economic benefit reward factor of the charging and discharging strategy, the reward factor of the demand guarantee of the electric vehicle trip, and the comprehensive reward function corresponding to each action is obtained.

[0031] Implement the update state vector S and input it into the trained DQN algorithm model, output the action selection probability distribution π(a|s) and the action a selected by the ε-greedy strategy according to the network, and control the charging and discharging behavior of the electric vehicle.

[0032] In some embodiments, the charging and discharging strategy is dynamically generated according to the power grid load prediction data and based on the DQN algorithm model, and the mature DQN algorithm model acquisition step includes:

[0033] The current electricity price, the remaining battery capacity, the number and distribution of electric vehicles accessing, and the information of the power grid carrying capacity of different regions are constructed into an environment model, and the state vector S is represented as:

[0034] S=[L t , L t+1 , …, L t+n , P t , E r1 , E r2 , …, E rm , N ev , C a1 , …, C ak ]

[0035] L t represents the current power grid load, and is assigned the power grid load of the first historical period, L t+i represents the predicted load of the future i time periods, and is assigned the predicted load of the i historical period, wherein the value of i is any one of 1 to n, P t represents the current electricity price, and is assigned the electricity price of the first historical period, E rj represents the remaining capacity of the jth electric vehicle, and is assigned the remaining capacity value of the jth electric vehicle in the first historical period, wherein the value of j is any one of 1 to m, N evThis represents the total number of electric vehicles connected to the power grid, assigned the value as the total number of electric vehicles connected to the power grid in the first historical period, C. al C represents the grid capacity-related indicators for the l-th region. ah This represents the power grid capacity-related index for the h-th region, where the value of h is any one from 1 to k;

[0036] Define a discrete action set A, which includes at least one action among fast charging, slow charging, pause charging, and battery discharge feeding back to the grid. Input the state vector S into the neural network and output the selection probability distribution π(a|s) for each action, where π is the probability distribution, a represents the action, and s represents the state.

[0037] The comprehensive reward function is obtained based on the power grid peak-valley difference adjustment reward factor, the charging and discharging strategy economic benefit reward factor, the electric vehicle travel demand guarantee reward factor, and the corresponding action for each historical period.

[0038] Multiple sets of state vectors S are input into the DQN algorithm model for training, resulting in a mature DQN algorithm model. It should be noted that the English name for the DQN algorithm model is Deep Q-Network, a reinforcement learning algorithm model based on deep learning.

[0039] In some embodiments, the reward function is derived based on the power grid peak-valley difference adjustment reward factor, the charging and discharging strategy economic benefit reward factor, the user satisfaction reward factor, and the comprehensive reward function corresponding to each action.

[0040] The power grid peak-valley difference adjustment reward factor includes the load peak-valley difference adjustment reward factor and the overload prevention penalty factor. The load peak-valley difference adjustment reward factor is expressed as follows:

[0041] r grid,tab =k1×(∣L peak -L valley | prev -∣L peak -L valley | new k1 is the weighting coefficient for the load peak-valley difference adjustment reward, used to adjust the size of the reward. L peak L represents the peak load of the power grid, that is, the load on the power grid during peak hours. valley This represents the off-peak load of the power grid, that is, the load on the power grid during off-peak hours, |L peak -L valley | prev |L represents the difference between peak and valley loads of the power grid before the action. peak -L valley | new This indicates the difference between peak and valley loads in the power grid after the action is taken;

[0042] The overload prevention penalty factor is expressed as:

[0043] r grid,verload =-k2×(L current -L safemax ), k2 is a penalty coefficient, L current为 is the current load, L safemax is the upper limit load of the safe load bearing;

[0044] The economic benefit reward factor of the charging and discharging strategy is expressed as:

[0045] r eco =k3×(E charged ×(P high -P low ) eff ),

[0046] k3 is an economic weight, E charged is the effective charging and discharging electric quantity, the charging is positive, the discharging is negative, (P high -P low ) eff is the effective price difference considering the charging and discharging efficiency loss, P high represents the electricity price in the peak period, and P low represents the electricity price in the valley period;

[0047] The user satisfaction reward factor is expressed as:

[0048] r use r=k4×(E rtarget -|E rtarget -E rj |), k4 is a user satisfaction weight, used to adjust the size of the reward, E rtarget is the target remaining electric quantity, i.e., the required electric quantity estimated by the user based on daily travel habits, E rj is the actual remaining electric quantity, i.e., the actual electric quantity of the electric vehicle at present;

[0049] The comprehensive reward function can be expressed as:

[0050] R(s,a)=ω1×r grid,tab +ω2×r grid,overload +ω3×r eco +ω4×r user , wherein ω1, ω2, ω3, and ω4 are dimension weights.

[0051] In a second aspect, the present application provides an electric vehicle two-way charging power grid load intelligent scheduling system based on artificial intelligence, comprising:

[0052] The data acquisition and processing module is configured to periodically acquire power grid load data of a preset power consumption area, and perform data cleaning, feature extraction, and data fusion on the power grid load data to generate a fusion data set, wherein the power grid load data comprises first load data, electric vehicle charging demand data, user behavior data, and weather data.

[0053] The load prediction module is configured to process the fusion data set by using a first algorithm model to obtain load prediction data of the power grid.

[0054] The intelligent scheduling optimization module is configured to dynamically generate a charging and discharging strategy based on the load prediction data of the power grid and a DQN algorithm.

[0055] The bidirectional charging management module is configured to supply power from the power grid to the electric vehicle battery or supply power from the electric vehicle battery to the power grid according to the charging and discharging strategy.

[0056] In some embodiments, the data acquisition and processing module comprises:

[0057] The data acquisition module is configured to periodically acquire and transmit the power grid load data.

[0058] The data storage module is configured to receive and store the power grid load data acquired by the data acquisition module.

[0059] The data calling module is configured to call the data in the data storage module.

[0060] The data processing module is configured to receive the power grid load data called by the first data calling module, and perform data cleaning, feature extraction, and data fusion on the power grid load data to generate a fusion data set.

[0061] In a third aspect, the present application provides an electronic device, which comprises:

[0062] at least one processor; and

[0063] a memory in communication with the at least one processor; wherein

[0064] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any of the embodiments of the artificial intelligence-based intelligent scheduling method for the bidirectional charging of the electric vehicle power grid load according to the first aspect.

[0065] In a fourth aspect, the present application provides a computer-readable storage medium storing at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement any of the embodiments of the artificial intelligence-based intelligent scheduling method for the bidirectional charging of the electric vehicle power grid load according to the first aspect.

[0066] For optimizing power grid load management, balancing electricity peak and valley, and improving the participation of electric vehicles in the power grid.

[0067] Compared with the prior art, the beneficial effects of the present application are that the electric vehicle bidirectional charging power grid load intelligent scheduling method based on artificial intelligence can integrate various types of data, including load data, real-time load data, electric vehicle charging demand data, user behavior data, and weather data, and through data cleaning, feature extraction, and data fusion technologies, more rich and valuable feature information can be mined, so that the change law and influencing factors of the power grid load can be more comprehensively reflected; considering the influence of load data, real-time load data, electric vehicle charging demand data, user behavior data, and weather data factors on the power grid load, the accuracy of power grid load data prediction can be greatly improved, more personalized and refined power supply strategies can be formulated according to different users and different scenarios, the needs of users can be better met, and the economy and reliability of power grid operation can be improved; during the peak period of power grid load, the charging and discharging state of electric vehicles can be adjusted in time to reduce the load impact on the power grid; in the event of power grid failure or emergency, the energy storage function of the electric vehicle battery can be used to provide temporary support for the power grid, and the emergency response capability and power supply reliability of the power grid can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0069] Figure 1 is a logic block diagram of the electric vehicle bidirectional charging power grid load intelligent scheduling method based on artificial intelligence;

[0070] Figure 2 is a schematic diagram of the electric vehicle bidirectional charging power grid load intelligent scheduling system based on artificial intelligence;

[0071] Figure 3 is another schematic diagram of the electric vehicle bidirectional charging power grid load intelligent scheduling system based on artificial intelligence;

[0072] Figure 4 is an electronic schematic diagram; DETAILED DESCRIPTION

[0073] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0074] It should be noted that the terms "comprising" and "having" and any variations thereof in the embodiments of the present application and the drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally further include steps or units not listed or can optionally further include other steps or units inherent to the process, method, product or device.

[0075] It can be understood that the terms "first", "second" and the like used in the present application can be used herein to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from another element.

[0076] The embodiments of the present application disclose an artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method, system, device, and storage medium, which are described in detail below.

[0077] Reference Figure 2 An artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling system includes a data acquisition and processing module 1, a load prediction module 2, an intelligent scheduling and optimization module 3, and a bidirectional charging management module 4. The data acquisition and processing module is used to periodically acquire power grid load data, and to perform data cleaning, feature extraction, and data fusion to generate a fusion data set, first load data, electric vehicle charging demand data, user behavior data, and weather data. The load prediction module 2 processes the fusion data set through a first algorithm model to obtain load prediction data of the power grid. The intelligent scheduling and optimization module generates a charging and discharging strategy based on the load prediction data of the power grid and a DQN algorithm. The bidirectional charging management module 4 is used to supply power to the electric vehicle battery or to supply power from the electric vehicle battery to the power grid according to the charging and discharging strategy.

[0078] Reference Figure 3In some embodiments, in an artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling system, the data acquisition and processing module 1 includes a data acquisition module 11, a data storage module 12, a first data calling module 13, and a data processing module 14. The data acquisition module 11 is used to periodically acquire and transmit power grid load data; the data storage module 12 is used to receive and store the power grid load data acquired by the data acquisition module 11; the first data calling module 13 is used to call the data in the data storage module 12; and the data processing module 14 is used to receive the power grid load data called by the first data calling module, and to perform data cleaning, feature extraction, and data fusion to generate a fusion data set.

[0079] Reference Figure 1 In some embodiments, an artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method can be implemented based on an electric vehicle bidirectional charging power grid load intelligent scheduling system. The specific method is as follows:

[0080] S100: periodically acquiring power grid load data of a preset time period in a preset power consumption area, the power grid load data including load data, electric vehicle charging demand data, user behavior data, and weather data;

[0081] S200: performing data cleaning, feature extraction, and data fusion on the power grid load data to generate a fusion data set;

[0082] S300: a first algorithm model processing the fusion data set to obtain power grid load prediction data;

[0083] S400: dynamically generating a charging and discharging strategy based on the power grid load prediction data and the DQN algorithm;

[0084] S500: scheduling power supply from the power grid to the electric vehicle battery or scheduling power supply from the electric vehicle battery to the power grid according to the charging and discharging strategy.

[0085] The load data includes, but is not limited to, voltage data, current data, power data, and power consumption period data.

[0086] The preset power consumption area is a specific power supply geographical division predefined in the power system operation and planning. The preset time period refers to a specific time interval for collecting power grid load data before the entire intelligent scheduling system is running, which is set in advance according to system design goals, power grid operation characteristics, user power consumption habits, and electric vehicle charging demand, etc. This time interval can be fixed, for example, with a 10-15 minute collection period, data collection from 9 am to 9 pm every day. In this time range, the power grid load changes more complexly and the electric vehicle charging activities are more frequent, which can obtain sufficient information for subsequent analysis and decision-making.

[0087] Specifically, since the electric energy is distributed to each distribution network line and user terminal after being output from the transformer substation, at the outlet line of each voltage level of the transformer substation in the preset power consumption area, a voltage transformer is installed to monitor the voltage to obtain voltage data, a current transformer is installed to detect current data, or a high-precision power meter is installed to obtain power data. These voltage data can reflect the voltage level of the power grid in the area responsible by the transformer substation.

[0088] The distribution network further distributes electric energy to each user area. Voltage transformers are installed at the sectionalizing switch positions of the distribution network lines, which can monitor the voltage of different lines in the preset power consumption area to timely find the voltage drop problem on the line, especially in the case of long-distance distribution or uneven load distribution, the specific line section where the voltage is too low or too high can be located. At the same time, current transformers and high-precision power detectors installed at the sectionalizing switch positions of the distribution network lines can obtain corresponding current data and power data.

[0089] Voltage transformers, current transformers, and high-precision power meters can also be installed at the sectionalizing switch positions of the distribution network lines to monitor the voltage, current, and power of different lines, which can be used as an important basis for developing distribution strategies, helping load balancing, guiding voltage regulation, influencing distributed power access control and optimization, and ensuring good operation of the distribution system in the preset power consumption area. Voltage detection, current detection, and power detection at the user terminal meter can more accurately measure the actual voltage, actual current, and actual power received by the user.

[0090] Power consumption period data can be obtained through smart meters, which divide a day into multiple periods, such as peak time, flat time, and valley time. According to the pre-set period division rules, the power consumption of each period is measured. These period divisions can be customized according to local electricity price policies and power grid operation characteristics. The clock module inside the meter ensures the accuracy of the time, and at the end of each period, the power consumption of that period is automatically recorded and stored in the meter's storage unit along with the time stamp. Then, through the communication interface, these power consumption data with time markers are regularly uploaded to the data management system of the power company for analyzing user power consumption behavior patterns, developing time-of-use pricing strategies, and power grid load forecasting. Time-of-use pricing policies will guide users to change their charging behavior. If a low valley electricity price is set, many users will adjust their charging time to the low valley period, which will increase the load of the power grid during that period, while the load during the peak period may decrease. This requires re-evaluation of the load situation in different periods during load forecasting to more accurately arrange power generation plans and power grid scheduling.

[0091] It should be noted that by periodically collecting power grid load data of the preset power consumption area in a specific time period, including load, electric vehicle charging demand, user behavior, weather, etc., after data cleaning, feature extraction and fusion, the first algorithm model is used to process to obtain prediction data, and then the DQN algorithm is used to generate charging and discharging strategies to realize scheduling, which can comprehensively consider influencing factors and more accurately grasp the load change trend, providing a reliable basis for subsequent decision-making; the predicted and generated strategies can be used to scientifically allocate the power between the power grid and the electric vehicle battery, realize peak clipping and valley filling, optimize resource utilization, ensure stable operation of the power grid, improve the ability of the power grid to respond to complex power consumption and electric vehicle charging demand, and enhance the economic efficiency and safety of overall operation.

[0092] In some embodiments, in an artificial intelligence-based intelligent scheduling method for electric vehicle bidirectional charging power grid load, the power data includes active power data, reactive power data, and power factor data.

[0093] Specifically, active power is calculated by measuring voltage and current effective values, reflecting actual power consumption; reactive power is related to inductive or capacitive elements, affecting voltage level; power factor is the cosine value of the phase difference between voltage and current, measuring power system efficiency.

[0094] It should be noted that in the artificial intelligence-based intelligent scheduling method for electric vehicle bidirectional charging power grid load, active power data reflects actual power consumption, reactive power data is related to the influence of inductive or capacitive elements on voltage level, and power factor data measures power system efficiency. These power data help accurately predict power grid load. Accurate prediction can know the load change trend in advance, provide a basis for load scheduling, so that the power grid can reasonably arrange power generation plans according to active and reactive power demand at different times, optimize the proportion of traditional energy and new energy generation, and allocate resources to ensure stable power supply before the load peak. In addition, by adjusting the power factor, the power quality can be improved and the loss can be reduced. During load scheduling, the power grid operating state can be monitored in real time based on power data, abnormalities can be found and adjusted in time, peak clipping and valley filling can be realized, and the economic efficiency, safety and stability of power grid operation can be improved. The power interaction between electric vehicles and the power grid can be effectively coordinated.

[0095] In some embodiments, the weather data includes at least one of temperature data, humidity data, wind speed data, rainfall data, snowfall data, lightning data, solar intensity data, haze data, and sand data.

[0096] Specifically, temperature data includes but is not limited to daily maximum temperature, daily minimum temperature, average temperature, temperature change rate. Temperature data can affect the use of refrigeration equipment such as air conditioning, high temperature in summer makes air conditioning and other refrigeration equipment use a lot, low temperature in winter makes heating equipment use increases, all of which increase the load of power grid. Humidity data includes but is not limited to relative humidity, absolute humidity. Increased humidity makes people feel hotter, so they use fans, air conditioners and other equipment more frequently, increasing the load of power grid. Wind speed data includes but is not limited to average wind speed, maximum wind speed, time distribution of wind speed. Rainfall data includes but is not limited to rainfall amount, rainfall duration, rainfall intensity. Snowfall data includes but is not limited to snowfall data, snowfall duration, snow depth data. Lightning data includes but is not limited to time, frequency, intensity, location data of lightning occurrence. Specifically, solar radiation intensity data includes but is not limited to sunshine duration, average solar radiation intensity, solar radiation change data at different times. Fog and haze data includes but is not limited to concentration, duration, coverage of fog and haze. Dust data includes but is not limited to intensity, duration, impact range of dust.

[0097] Temperature data, humidity data, wind speed data, rainfall data can be obtained by thermometer of meteorological observation station, or by satellite remote sensing, numerical weather prediction model.

[0098] It should be noted that the use of various weather data such as temperature, humidity, wind speed, rainfall, snowfall, lightning, solar radiation intensity, fog and haze, and dust in power grid load forecasting can consider influencing factors from multiple dimensions, such as temperature and humidity affecting refrigeration and heating equipment power consumption, wind speed affecting renewable energy power generation, lightning affecting other factors, avoiding single factor limitation, and capturing complex nonlinear and dynamic change relationship between weather and load, improving prediction accuracy; In power grid load dispatching, power generation resource allocation can be planned in advance according to accurate prediction, extreme weather conditions can be flexibly responded to, new energy consumption and synergy can be promoted, extreme weather conditions can be responded to, power grid safety and stability can be enhanced, including optimizing equipment maintenance, adjusting voltage and frequency, etc., to ensure reliable operation of power grid.

[0099] In some embodiments, electric vehicle charging demand data includes user basic information, user basic information includes user number data, user distribution data, user type data, user travel habit data; vehicle information, vehicle information includes vehicle ownership data, vehicle battery parameter data, vehicle range data; charging behavior information, charging behavior information charging time data, charging location preference data, charging frequency data. Market and policy data, market and policy data includes market growth rate data, policy guidance data, wherein the market growth rate data includes load size change data and load space-time distribution change data, and the policy guidance data includes data related to user charging strategy change caused by electricity price policy.

[0100] The user number data can determine the overall size of the electric vehicle users. The user distribution data can determine the distribution of users in the geographical space. The user distribution data can be statistically subdivided according to the administrative division of the city, such as districts, counties, or according to the functional areas such as commercial areas, residential areas, industrial areas, science and technology parks, etc.

[0101] The user type data can be divided into individual users, enterprise users, and operating vehicle users. Different types of users have significant differences in charging demand and behavior patterns. The charging time of individual users is usually flexible, which can be concentrated after work at home or when going out for business, shopping, and leisure; enterprise users may consider setting charging piles in the company parking lot to meet the charging needs of employees during work hours, and enterprise users may have specific requirements and specifications for the management and use of charging facilities; operating vehicle users need to charge frequently to ensure the continuous operation of the vehicle, and the charging time and place are relatively fixed, and the charging speed and efficiency are required to be high. The user travel habit data includes but is not limited to the user's daily travel path, travel distance, and travel time distribution. Through the mining and analysis of a large amount of user travel data, the user's travel regularity can be understood.

[0102] The vehicle battery parameter data includes but is not limited to the capacity of the battery, the charging rate, the type of the battery, the state of health of the battery, etc. In some embodiments, the type of the battery can be a lithium-ion battery or a lithium iron phosphate battery. The battery capacity determines the maximum mileage that the vehicle can travel after charging, and different capacity batteries have different charging needs and times. The charging rate affects the charging speed, and the battery with high charging rate can charge more power in a short time. Determining the type of the battery helps to determine the appropriate charging technology and charging strategy. The state of health of the battery is also important for evaluating the charging efficiency and the change of the vehicle's range, and as the use time of the battery increases and the number of charging and discharging increases, the state of health of the battery will gradually decrease, which may cause problems such as prolonged charging time and shortened range.

[0103] The vehicle range data is closely related to the battery capacity and the energy efficiency of the vehicle, and is an important indicator of how far the electric vehicle can travel after charging. The range data not only affects the user's travel planning and charging decision.

[0104] The charging time data includes recording the specific time points and duration of the user charging at different time periods within 24 hours or a week. Through the analysis of the charging time data, the obvious charging peak and valley periods can be found.

[0105] The charging location preference data reflects the user's tendency when choosing a charging location. The user can decide the charging location according to his own travel destination, residence location, work location, and the convenience, charging speed, charging cost, and other factors of the charging facility. The charging location can be one of the travel destination, residence location, and work location.

[0106] The charging frequency data reflects the number of times the user charges the vehicle within a certain time period. The charging period can be a week, a month, or other custom-defined periods according to needs. The charging frequency is related to the user's travel mileage, vehicle range, and charging habits. For example, for users with short daily travel mileage and long vehicle range, their charging frequency may be relatively low. For users of operating vehicles, due to the frequent use of vehicles, they may need to charge multiple times a day.

[0107] In some embodiments, the user behavior data includes timestamp data, time period data, instantaneous power data, average power data, device classification data, device quantity data, user category data, and geographic location data.

[0108] Specifically, the timestamp data is presented in the format of year, month, day, hour, minute, and second, representing the time label corresponding to an action. It clearly tracks when the user performed an operation, such as a laundry worker starting work at 19:30:15 on October 1, 2024. The timestamp data can be used to track the order of user operations and the chronological relationship of events, and to determine the exact time of a fault occurrence. The time period data represents the repeated regularity of user behavior within a certain time range. The instantaneous power data is the power consumed by the electrical equipment at a specific moment. The average power data is the average value of the power consumed by the electrical equipment within a specific time interval. The device classification data includes but is not limited to device category, model, and function. The user category data includes but is not limited to the number and scale of residential users, industrial users, commercial users, and agricultural users.

[0109] Further, the market growth rate data refers to the change in the total load of the power grid in the market environment of electric vehicle charging interacting with the power grid within a predetermined time period. It can be measured by statistics of the total charging power of all electric vehicles connected to the power grid and the total power of other conventional electrical equipment in different time periods, such as hours, days, months, and years. For example, as the number of electric vehicles increases, the total load of the power grid may show a gradual upward trend; or in some specific time periods, such as night low-price electricity, a large number of electric vehicles charge at the same time, causing the load size to jump significantly in that period. The change in load size data is a key indicator for evaluating market growth rate, which helps the power grid company to plan power grid expansion and upgrade in advance to cope with the growing demand for electricity. At the same time, it also provides a basis for power generation companies to adjust power generation plans and reasonably arrange energy resources to ensure the stability and reliability of power supply.

[0110] Further, the load space-time distribution change data refers to the distribution change of electric vehicle charging load and other electrical load in time and space. From the time dimension, it includes the load distribution in different time periods within a day, such as peak, flat, and valley load distribution, as well as seasonal load fluctuations within a week, a month, or a year. From the spatial dimension, it involves the load distribution difference in different geographical areas, such as cities, suburbs, and different power supply zones. By understanding the distribution rules of load in time and space, the power grid company can develop more accurate time-of-use pricing strategies to guide users to charge during the load valley period, thereby achieving the peak shaving and valley filling of the power grid. Load space-time distribution change data includes time dimension data and spatial dimension data. Time dimension data can be recorded by smart meters and power grid data acquisition systems at each time point, and then analyzed statistically. Spatial dimension data needs to be associated with the geographical location information of each user and the load data to draw a load spatial distribution map through the geographic information system. At the same time, monitoring and statistics of the use of charging facilities in different areas can also provide data support for the analysis of load spatial distribution.

[0111] Further, the policy-oriented data refers to data related to the change of user charging strategy caused by electricity price policy. When the government or power regulatory department introduces electricity price policy, such as time-of-use electricity price, peak-valley electricity price, and step electricity price, users may adjust their charging time, charging location, and charging amount, etc. For example, when implementing peak-valley electricity price policy, users may transfer the charging time of electric vehicles from peak time to valley time or charge in areas with lower electricity price in order to save electricity cost. These data of user charging strategy change can reflect the guiding effect of policy on market behavior. By analyzing the change of user charging strategy, it can be understood whether the policy has achieved the expected goal, such as relieving the pressure of peak load of power grid and promoting the rational use of energy, etc. It is helpful for them to adjust the operation strategy and service content according to the change of user behavior. For example, in the valley time period where the user charging demand is concentrated, the maintenance and operation cost of charging facilities can be appropriately reduced to improve service efficiency; or promotion activities and value-added services can be increased in the charging location preferred by users. The policy-oriented data can record the charging behavior data of users, including charging time, charging location, and charging amount, etc. By comparing the change of these data before and after the implementation of electricity price policy, the change of user charging strategy can be analyzed. At the same time, combined with user survey and market feedback, information such as user's awareness, acceptance degree of electricity price policy, and reasons for the change of charging behavior can be obtained, so as to more comprehensively evaluate the influence of policy-oriented on user charging strategy.

[0112] In some embodiments, the power grid load data is subjected to data cleaning, feature extraction, and data fusion to generate a fused data set, including:

[0113] S2001: The data cleaning includes supplementing the missing power grid load data by mean interpolation or linear interpolation method, calculating the standard deviation and mean value of the corresponding power grid load data, regarding the data points exceeding the mean value ± 2 times the standard deviation range as abnormal values, adjusting the abnormal values by analyzing the trend of adjacent normal data points, identifying and deleting duplicate values by comparing the fields of data records;

[0114] S2002: The feature extraction includes calculating at least one of the maximum value, minimum value, average value, and standard deviation of the power grid load data, calculating the average value of the power grid load data in the first preset time period to obtain the moving average feature data, calculating the moving autocorrelation coefficient feature data packet of the first power grid load data packet at the starting time point and the second power grid load data packet at the terminal time point in the preset second time period, performing fast Fourier transform on the power grid load data to obtain the Fourier transform feature value, and calculating the change rate data based on the power grid load data, including:

[0115] S2003: The data fusion includes fusing power grid load data, charging demand data, user behavior data, and weather data to form a fused data set.

[0116] Specifically, in the data cleaning process, voltage data or current data is arranged in corresponding data columns in time sequence, respectively. For missing data in the data columns, mean interpolation or linear interpolation method is used for supplement. Mean interpolation is to calculate the mean value of the column where the data is located, and replace the missing value with the mean value and supplement into the corresponding data column. Linear interpolation method is to estimate the missing value by constructing a linear function according to the known data points before and after the missing data. Further, if the current data is missing at a certain time, and the current value at the previous time is I1, and the current value at the next time is I2, and the time interval is Δt, and the time interval from the previous time to the missing time is t, then the missing current value I = I1 + t(I2-I1) / Δt. The processing method of other missing power grid load data is the same. For the missing power grid load data with linear law, linear interpolation method is used, and for the missing data without linear law, mean interpolation method is selected. For the supplemented power grid load data, the corresponding standard deviation and mean value are calculated, the data points exceeding the range of ±2 times the standard deviation of the mean value are regarded as abnormal values, and the abnormal values are adjusted by analyzing the trend of adjacent normal data points, and the repeated values are identified and deleted by comparing the fields of the data records.

[0117] Specifically, in the feature extraction process, the maximum value, minimum value, average value, standard deviation and the like of the power grid load data are calculated. For example, for the voltage data within a day, the maximum value and minimum value are found, and the average value and standard deviation are calculated. The dispersion degree and overall level of the voltage data can be obtained. The average value of the power grid load data in the first preset time period, such as the past 1 hour, is calculated to obtain the moving average feature data.

[0118] Further, it is assumed that data is collected every m minutes, and k = 60 / m data points L1, L2, …, L k are collected in the past 1 hour. Then the moving average feature value M A is calculated. A The calculation formula is: M k = (L1+L2+…+L p ) / k. The moving average value feature can smooth short-term fluctuations and highlight long-term trends. The moving autocorrelation coefficient feature data packet of the first power grid load data packet at the starting time point and the second power grid load data packet at the terminal time point in the second preset time period, such as the past 3 hours, is calculated. It is assumed that the starting time point data packet is P1 = (x1, x2, …, xp), and the terminal time point data packet is P2 = (y1, y2, …, yp). The calculation formula of the moving autocorrelation coefficient is: r = (x1*y1+x2*y2+…+xp*yp) / (p*p). q ave p ave j1+j2+…+jp) / (p*p).​​​p The calculation formula of the autocorrelation coefficient r is:

[0119] r = [(x1-x ave )*(y1-y ave )+(x2-x ave )*(y2-y ave )+…+(x p -x ave )*(y p -x ave )] / {[(x1-x ave )+(x2-x ave )+…+(x p -x ave )]*[(y1-y ave )+(y2-y ave )+…+(y p -x ave )] 1 / 2

[0120] Where x ave and y ave are the mean values of P1 and P2, respectively. By calculating the autocorrelation coefficient, the correlation between data at different time periods can be understood, and the periodic or trend changes of the data can be determined.

[0121] Further, the power grid load data is subjected to fast Fourier transform to obtain Fourier transform characteristic values. Fourier transform can convert time domain data into frequency domain data, thereby analyzing different frequency components in the data. For example, after Fourier transform of the current data, the amplitude and phase information of different frequency harmonics in the current signal can be obtained, which is helpful to find harmonic pollution or other frequency-related problems in the power grid.

[0122] Further, a change rate data packet is calculated based on the power grid load data. For two consecutive data points D1 and D2, the calculation formula of the change rate CR is: CR = (D2-D1) / D1, and the change rate feature can reflect the change speed of the power grid load data. Each group of data in the power grid load data can be brought into the calculation one by one to obtain the corresponding change rate data packet.

[0123] In some embodiments, the first algorithm model processes the fusion data set to obtain power grid load prediction data, including:

[0124] S3001: converting the fusion data set into matrix data, normalizing the matrix data, and dividing the data into a training set, a validation set, and a test set;

[0125] S3002: The first algorithm model includes an LSTM model and a GRU model, the training set is input into the LSTM model and the GRU model for pre-training, the first hidden state set and the second hidden state set are obtained based on the respective independent loss function and the optimizer, and the first hidden state and the second hidden state set are fused to obtain the fusion hidden state set, and after the fusion state set is input into the shared output layer, the power grid load prediction data is obtained.

[0126] The fusion data set is converted into matrix data, which enables the model to simultaneously process multiple factor correlations, is beneficial to mining the comprehensive influence on the load, and can comprehensively and accurately divide the electricity price period on the power grid strategy; the normalization processing makes the model treat each feature fairly, improves the load prediction accuracy, and also provides a reliable basis for the power grid dispatching strategy; reasonable data division ensures effective training and evaluation of the model, and its accurate prediction helps to accurately formulate the power generation plan and other strategies. In the S3002 step, the LSTM model captures the long-term dependence of the load by means of the gating mechanism, the GRU model has a simple structure and can quickly learn short-term features, and the pre-training of the two models is important in long-term planning and short-term dispatching strategy formulation; the hidden state set fusion integrates the advantages of both, which can more comprehensively describe the change law in load prediction, and thus provide support for fine management strategies such as power grid and electric vehicle interaction, effectively improving the accuracy of power grid strategies and the accuracy of load prediction. Through the joint training of the LSTM model and the GRU model, the most relevant data can be selected to facilitate more accurate subsequent prediction of the power grid load, thereby improving the accuracy of power grid dispatching, and thus improving the economy, reliability and adaptability of power grid operation; at the same time, the use of the common output layer can reduce the data processing path and improve the data processing efficiency.

[0127] In some embodiments, according to the power grid load prediction data, and based on the DQN algorithm, the charging and discharging strategy is dynamically generated, including:

[0128] S3001: An environment model is constructed by using the current electricity price, the remaining battery capacity, the number and distribution of electric vehicles connected, the power grid carrying capacity of different regions and other information, and the state vector S is represented as:

[0129] S = [L t , L t+1 , …, L t+n , P t , E r1 , E r2 , …, E rm , N ev , C a1 , …, C ak ]

[0130] L t represents the current power grid load, L t+iP represents the predicted load of the future i time periods t E represents the current electricity price rj N represents the remaining power of the jth electric vehicle ev C represents the total number of electric vehicles accessing the power grid al The grid capacity related index of the lth region is represented.

[0131] S3002: Define a discrete action set A, the action set A includes at least one of fast charging, slow charging, pausing charging, and battery discharging to feed back to the grid, input the state vector S into the neural network, and output the selection probability distribution π(a|s) of each action, the π probability distribution, a represents action, and s represents state;

[0132] S3003: Adjust the reward factor according to the grid peak-valley difference, the charging and discharging strategy economic benefit reward factor, the electric vehicle travel demand guarantee reward factor, and the comprehensive reward function corresponding to each action;

[0133] S3004: Real-time update of the state vector S and input into the DQN network;

[0134] S3005: Real-time update of the state vector S and input into the trained DQN network, and the action selection probability distribution π(a|s) and the action a selected by the ε-greedy strategy according to the network output are used to control the charging and discharging behavior of the electric vehicle.

[0135] In some embodiments, in the step of obtaining a comprehensive reward function corresponding to each action according to the grid peak-valley difference adjustment reward factor, the charging and discharging strategy economic benefit reward factor, the user satisfaction reward factor, and the comprehensive reward function corresponding to each action,

[0136] The grid peak-valley difference adjustment reward factor includes a load peak-valley difference adjustment reward factor and an overload prevention penalty factor, and the load peak-valley difference adjustment reward factor is represented as:

[0137] r grid,tab =k1×(∣L peak -L valley ∣ prev -∣L peak -L valley ∣ new ), k1 is the weight coefficient of the load peak-valley difference adjustment reward, used to adjust the size of the reward, L peak represents the peak load of the grid, i.e. the load amount of the grid at the peak period, L valley represents the valley load of the grid, i.e. the load amount of the grid at the valley period, |L peak -L valley ∣ prev represents the difference between the peak and valley loads of the grid before the action, and |L peak -Lvalley ∣ new represents the difference between the peak and valley loads of the power grid after the action;

[0138] The overload prevention penalty factor is represented as:

[0139] r grid,verload = -k2 x (L current - L safemax ), k2 is a penalty coefficient, L current为 is the current load, L safemax is the safe carrying upper limit load;

[0140] The charging and discharging strategy economic benefit reward factor is represented as:

[0141] r eco = k3 x (E charged x (P high - P low ) eff ),

[0142] k3 is an economic weight, E charged is the effective charging and discharging electric quantity, positive for charging and negative for discharging, (P high - P low ) eff is the effective price difference considering the charging and discharging efficiency loss, P hig represents the price of the peak period, which is the period of the highest demand for electricity in a day, and the price will be relatively high, P low represents the price of the valley period, which is the period of the lowest demand for electricity in a day, and the price will be relatively low;

[0143] The user satisfaction reward factor is represented as:

[0144] r use r = k4 x (E rtarget - |E rtarget - E rj |), k4 is a user satisfaction weight, used to adjust the size of the reward, E rtarget is the target remaining electric quantity, i.e., the required electric quantity estimated by the user based on daily travel habits, E rj is the actual remaining electric quantity, i.e., the actual electric quantity of the electric vehicle at present;

[0145] The comprehensive reward function can be represented as:

[0146] R(s, a) = ω1 x r grid,tab + ω2 x r grid,overload + ω3 x r eco + ω4 x r user , wherein ω1, ω2, ω3, and ω4 are the weights of each

[0147] dimension.

[0148] In some embodiments, the artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method dynamically generates a charging and discharging strategy based on power grid load prediction data and a DQN algorithm model. The mature DQN algorithm model acquisition step includes:

[0149] An environment model is constructed with current electricity price, battery remaining capacity, number and distribution of electric vehicles connected, power grid carrying capacity in different regions, and other information. The state vector S is represented as:

[0150] S = [L t , L t+1 , …, L t+n , P t , E r1 , E r2 , …, E rm , N ev , C a1 , …, C ak ]

[0151] L t represents the current power grid load, which is assigned to the power grid load of the first historical period, L t+i represents the predicted load of the future i time periods, which is assigned to the predicted load of the i-th historical period, where i is any value from 1 to n, P t represents the current electricity price, which is assigned to the electricity price of the first historical period, E rj represents the remaining capacity of the jth electric vehicle, which is assigned to the remaining capacity value of the jth electric vehicle in the first historical period, where j is any value from 1 to m, N ev represents the total number of electric vehicles connected to the power grid, which is assigned to the total number of electric vehicles connected to the power grid in the first historical period, C al represents the power grid capacity-related indicators of the lth region, C ah represents the power grid capacity-related indicators of the hth region, where h is any value from 1 to k.

[0152] A discrete action set A is defined, which includes at least one of fast charging, slow charging, pausing charging, and battery discharging to feed back to the power grid. The state vector S is input into a neural network, and the selection probability distribution π(a|s) of each action is output. The π probability distribution represents action (a) and state (s).

[0153] The comprehensive reward function is obtained according to the power grid peak-valley difference adjustment reward factor of the first historical period, the economic benefit reward factor of the charging and discharging strategy, the electric vehicle travel demand guarantee reward factor, and each action.

[0154] The plurality of groups of state vectors S are input into the DQN algorithm model for training, to obtain a mature DQN algorithm model.

[0155] It should be noted that the state vector S is formed by constructing an environment model containing current and future power grid load, electricity price, and residual electricity of electric vehicles, and the like, and defining a related discrete action set A, the parameters are assigned values based on the first historical period data, the comprehensive reward function is constructed according to a plurality of reward factors, and the DQN algorithm model is trained to maturity by using the plurality of groups of state vectors S. In terms of accurate prediction of power grid load, it can deeply mine various associated information in historical data, fully consider the comprehensive influence of electric vehicle related factors and power grid characteristics on load change, make the prediction model more suitable for actual complex scenarios, thereby improving the accuracy and reliability of prediction, and accurately grasping the dynamic change trend of load. In terms of power grid load dispatching, the mature DQN algorithm model can intelligently choose actions such as fast charging, discharging feedback, and the like under different power grid states according to the accurate prediction result, reasonably guide the charging and discharging behavior of electric vehicles, effectively balance power supply and demand, realize peak clipping and valley filling, optimize power grid resource allocation, improve the flexibility and stability of the power grid in response to different load conditions, enhance the economy and safety of power grid operation, and ensure the efficient and orderly operation of the power system and promote the benign interaction between electric vehicles and the power grid.

[0156] Reference Figure 4 In some embodiments, the artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent dispatching method can be built into the electronic device 5, which includes at least one processor 6; and a memory 7 connected in communication with the at least one processor 6; wherein the memory 7 stores instructions executable by the at least one processor 6, and the instructions are executed by the at least one processor 6 to enable the at least one processor 6 to perform any of the artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent dispatching methods in the present application.

[0157] In some embodiments, a computer-readable storage medium stores at least one instruction, which is executed by the processor 6 in the electronic device 5 to implement any of the artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent dispatching methods in the present application.

[0158] Those skilled in the art can understand that all or part of the steps of various methods of the above embodiments can be completed by instructing the relevant hardware by a program, and the program can be stored in a computer readable storage medium, including Read-Only Memory (ROM), Random Access Memory (RAM), Programmable Read-only Memory (PROM), Erasable Programmable Read Only Memory (EPROM), One-time Programmable Read-Only Memory (OTPROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM) or other optical disk memories, magnetic disk memories, magnetic tape memories, or any other computer readable medium capable of carrying or storing data.

[0159] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the electric vehicle bidirectional charging power grid load intelligent scheduling method and system based on artificial intelligence of the present application, but not to limit them; although the electric vehicle bidirectional charging power grid load intelligent scheduling method and system based on artificial intelligence of the present application are described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent scheduling of electric vehicle bidirectional charging power grid load based on artificial intelligence, characterized in that, The method comprises the following steps: Periodically collecting power grid load data in a preset time period of a preset power consumption area, the power grid load data including first load data, electric vehicle charging demand data, user behavior data, and weather data; Performing data cleaning, feature extraction, and data fusion on the power grid load data to generate a fused data set; A first algorithm model processes the fused data set to obtain power grid load prediction data; Based on the power grid load prediction data and a DQN algorithm, a charging and discharging strategy is dynamically generated; The charging and discharging strategy is used to schedule power supply from the power grid to the electric vehicle battery or from the electric vehicle battery to the power grid; The method of dynamically generating the charging and discharging strategy based on the power grid load prediction data and the DQN algorithm comprises the following steps: An environment model is constructed based on current electricity price, battery remaining capacity, number and distribution of electric vehicles connected, power grid carrying capacity of different areas, and other information, and a state vector S is represented as: S = [L t , L t+1 , …, L t+n , P t , E r1 , E r2 , …, E rm , N ev , C a1 , …, C ak ] L t represents the current grid load, L t+i represents the predicted load of the future i time periods, where i has any value from 1 to n, P t represents the current electricity price, E rj represents the remaining electricity of the jth electric vehicle, where j has any value from 1 to m, N ev represents the total number of electric vehicles connected to the grid, C ah represents the grid capacity related index of the hth region, where h has any value from 1 to k. A discrete action set A is defined, the action set A includes at least one of fast charging, slow charging, pausing charging, and battery discharging to feed back to the power grid, the state vector S is input into a neural network, and a selection probability distribution π(a|s) of each action is output, where π is a probability distribution, a represents an action, and s represents a state; A comprehensive reward function is obtained based on a reward factor adjusted according to a power grid peak-valley difference, an economic benefit reward factor of the charging and discharging strategy, an electric vehicle travel demand guarantee reward factor, and each action; The state vector S is updated and input into a trained DQN algorithm model, the network outputs an action selection probability distribution π(a|s) and an action a selected by an ε-greedy strategy, and the charging and discharging behavior of the electric vehicle is regulated. 2.The AI-based electric vehicle bidirectional charging grid load intelligent scheduling method of claim 1, wherein, The first load data includes voltage data, current data, power data, and power consumption time period data.

3. The method according to claim 2, wherein the power data includes active power data, reactive power data, and power factor data.

4. The method according to claim 1, wherein the weather data includes temperature data, humidity data, wind speed data, rainfall data, snowfall data, lightning data, sunshine intensity data, haze data, and sandstorm data. The electric vehicle charging demand data includes: User basic information, including user number data, user distribution data, user type data, and user travel habit data; 5. The method of claim 1, wherein the method is characterized by, Vehicle information, including vehicle population data, vehicle battery parameter data, and vehicle range data; Charging behavior information, including charging time data, charging location preference data, and charging frequency data; Market and policy data, including market growth rate data and policy guidance data, wherein the market growth rate data includes load size change data and load space-time distribution change data, and the policy guidance data includes data on changes in user charging strategies caused by electricity price policies. ​ ​ 6.The intelligent scheduling method for electric vehicle bidirectional charging power grid load based on artificial intelligence according to claim 1, characterized in that, The user behavior data includes timestamp data, time period data, instantaneous power data, average power data, device classification data, device quantity data, user category data, and geographic location data.

7. The intelligent scheduling method for electric vehicle bidirectional charging power grid load based on artificial intelligence according to any one of claims 1 to 6, characterized in that, The data cleaning, feature extraction and data fusion of the power grid load data in the fusion data set include: missing power grid load data is supplemented by mean interpolation or linear interpolation method, the standard deviation and mean value of the corresponding power grid load data are calculated, data points exceeding the mean value ± 2 times the standard deviation range are regarded as abnormal values, and the abnormal values are adjusted by analyzing the trend of adjacent normal data points, and the repeated values are identified and deleted by comparing the fields of data records; The feature extraction of the power grid load data in the fusion data set includes: at least one of the maximum value, the minimum value, the average value and the standard deviation of the power grid load data is calculated, the average value of the power grid load data in the first preset time period is calculated to obtain the moving average feature data, the moving autocorrelation coefficient feature data packet of the first power grid load data packet at the starting time point and the second power grid load data packet at the terminal time point in the preset second time period is calculated, the power grid load data is subjected to fast Fourier transform to obtain the Fourier transform feature value, and the change rate data packet is calculated based on the power grid load data base; The data fusion of the power grid load data in the fusion data set includes: The power grid load data and the charging demand data are fused, the user behavior data and the weather data are fused to form a fusion data set. 8.The AI-based electric vehicle bidirectional charging grid load intelligent scheduling method of claim 7, wherein, The processing of the fusion data set by the first algorithm model to obtain the power grid load prediction data includes: The fusion data set is converted into matrix data, the matrix data is normalized, and the data is divided into a training set, a validation set and a test set; The first algorithm model includes an LSTM model and a GRU model, the training set is input into the LSTM model and the GRU model for pre-training, a first hidden state set and a second hidden state set are obtained based on respective independent loss functions and optimizers, the first hidden state set and the second hidden state set are fused to obtain a fusion hidden state set, and the fusion state set is input into a shared output layer to obtain the power grid load prediction data. 9.The AI-based electric vehicle bidirectional charging grid load intelligent scheduling method of claim 8, wherein, The dynamic generation of the charging and discharging strategy based on the power grid load prediction data and the DQN algorithm model includes: The current electricity price, the remaining battery capacity, the number and distribution of electric vehicles, the power grid carrying capacity of different regions and other information are used to construct an environment model, and the state vector S is represented as: S = [L t , L t+1 , …, L t+n , P t , E r1 , E r2 , …, E rm , N ev , C a1 , …, C ak ] L t represents the current grid load, and is assigned the grid load of the first historical period, L t+i represents the predicted load of the future i time period, and is assigned the predicted load of the i-th historical period, where i is any value from 1 to n, P t represents the current electricity price, and is assigned the electricity price of the first historical period, E rj represents the remaining power of the j-th electric vehicle, and is assigned the remaining power value of the j-th electric vehicle of the first historical period, where j is any value from 1 to m, N ev represents the total number of electric vehicles connected to the grid, and is assigned the total number of electric vehicles connected to the grid of the first historical period, C al represents the grid capacity related index of the l-th region, C ah represents the grid capacity related index of the h-th region, where h is any value from 1 to k. Define a discrete action set A, the action set A includes at least one of fast charging, slow charging, pausing charging, and discharging the battery to feed back to the power grid, input the state vector S into a neural network, and output a selection probability distribution π(a|s) of each action, wherein π represents the probability distribution, a represents an action, and s represents a state; According to the power grid peak-valley difference adjustment reward factor, the charging and discharging strategy economic benefit reward factor, the electric vehicle travel demand guarantee reward factor and the comprehensive reward function corresponding to each action in the first historical period, Input multiple groups of state vectors S into a DQN algorithm model for training to obtain a mature DQN algorithm model. 10.The intelligent scheduling method for electric vehicle bidirectional charging power grid load based on artificial intelligence according to claim 8, characterized in that, According to the power grid peak-valley difference adjustment reward factor, the charging and discharging strategy economic benefit reward factor, the user satisfaction reward factor and the comprehensive reward function corresponding to each action, The power grid peak-valley difference adjustment reward factor includes a load peak-valley difference adjustment reward factor and an overload prevention penalty factor, and the load peak-valley difference adjustment reward factor is represented as: r grid,tab = k1 x (|L peak -L valley | prev - |L peak -L valley | new ), k1 is a weight coefficient of load peak-valley difference adjustment reward, used to adjust the size of the reward, L peak represents the peak load of the power grid, that is, the load amount of the power grid in the peak period, L valley represents the valley load of the power grid, that is, the load amount of the power grid in the valley period, |L peak -L valley | prev represents the difference between the peak-valley load of the power grid before the action, |L peak -L valley | new represents the difference between the peak-valley load of the power grid after the action; The overload prevention penalty factor is represented as: r grid,verload = -k2 x (L current -L safemax ), k2 is a penalty coefficient, L current is the current load, L safemax the upper load limit of the safe bearing The charging and discharging strategy economic benefit reward factor is represented as: r eco = k3 x (E charged x (P high - P low ) eff ), k3 is the economic weight, E charged is the effective charge and discharge power, the charge is positive, the discharge is negative, (P high - P low ) eff is the effective price difference considering the efficiency loss of charging and discharging, P hig represents the electricity price of peak period, P low represents the electricity price of valley period; The user satisfaction reward factor is represented as: r use r = k4 x (E rtarget - |E rtarget - E rj |), k4 is a user satisfaction weight used to adjust the size of the reward, E rtarget is the target remaining amount of electricity, i.e., the required amount of electricity estimated by the user based on daily travel habits, E rj is the actual remaining amount of electricity, i.e., the actual amount of electricity of the electric vehicle; The comprehensive reward function can be represented as: R(s, a) = ω1 x r grid,tab + ω2 x r grid,overload + ω3 x r eco + ω4 x r user where ω1, ω2, ω3, ω4 are the weights for each dimension.

11. An artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling system, characterized in that, including: A data acquisition and processing module is configured to periodically acquire power grid load data of a preset power consumption area, and perform data cleaning, feature extraction and data fusion on the power grid load data to generate a fusion data set, wherein the power grid load data includes first load data, electric vehicle charging demand data, user behavior data and weather data; A load prediction module is configured to process the fusion data set by using a first algorithm model to obtain load prediction data of the power grid; An intelligent scheduling optimization module is configured to dynamically generate a charging and discharging strategy based on the load prediction data of the power grid and the DQN algorithm; A bidirectional charging management module is configured to supply power to electric vehicle batteries or supply power from electric vehicle batteries to the power grid according to the charging and discharging strategy; The dynamic generation of the charging and discharging strategy based on the load prediction data of the power grid and the DQN algorithm includes: An environment model is constructed by using current electricity prices, remaining battery capacity, the number and distribution of electric vehicles connected, the power carrying capacity of different regions and other information, and a state vector S is represented as: S = [L t , L t+1 ,..., L t+n , P t , E r1 , E r2 ,..., E rm , N ev , C a1 ,..., C ak ] L t represents the current grid load, L t+i represents the predicted load of the future i time periods, where i has any value from 1 to n, P t represents the current electricity price, E rj represents the remaining electricity of the jth electric vehicle, where j has any value from 1 to m, N ev represents the total number of electric vehicles connected to the grid, C ah represents the grid capacity related index of the hth region, where h has any value from 1 to k. Define a discrete action set A, the action set A includes at least one of fast charging, slow charging, pausing charging, and discharging the battery to feed back to the power grid, input the state vector S into a neural network, and output a selection probability distribution π(a|s) of each action, wherein π represents the probability distribution, a represents an action, and s represents a state; According to the power grid peak-valley difference adjustment reward factor, the charging and discharging strategy economic benefit reward factor, the electric vehicle travel demand guarantee reward factor and the comprehensive reward function corresponding to each action in the first historical period, The updated state vector S is input into the trained DQN algorithm model, the network outputs an action selection probability distribution π(a|s) and an action a selected by an ε-greedy strategy, and the charging and discharging behavior of the electric vehicle is regulated.

12. The artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling system according to claim 11, characterized in that, The data acquisition and processing module includes: A data acquisition module is configured to periodically acquire and transmit the power grid load data; A data storage module is configured to receive and store the power grid load data collected by the data collection module. A first data calling module is configured to call the data in the data storage module. A data processing module is configured to receive the power grid load data called by the first data calling module, and perform data cleaning, feature extraction, and data fusion on the power grid load data to generate a fused data set.

13. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method according to any one of claims 1 to 10.

14. A computer-readable storage medium storing at least one instruction, wherein the at least one instruction causes a processor to perform operations comprising: The at least one instruction is executed by the processor in the electronic device to implement the artificial intelligence-based electric vehicle bidirectional charging power grid load intelligent scheduling method according to any one of claims 1 to 10.

Images