Intelligent charging scheduling method and system

Abstract

The invention discloses an intelligent charging scheduling method and system, and the method comprises the steps: collecting a real-time multi-source data set, building an LSTM charging demand prediction model based on an LSTM neural network, inputting the real-time multi-source data set into the LSTM charging demand prediction model for prediction, and defining a dynamic priority weight function according to the charging demand and the power grid schedulable margin. Generating a priority list; establishing a multi-target hierarchical optimization algorithm by utilizing MIP mixed integer programming and an NSGA-II genetic algorithm, and solving a charging scheduling scheme in a hierarchical manner by taking a priority list as a constraint based on the multi-target hierarchical optimization algorithm; active peak load shifting is performed by using a grading response strategy, the regional peak load is reduced, the capacity expansion investment of the transformer is delayed, and the service life of equipment is prolonged. And in combination with the V2G reverse power supply capability, electric vehicle energy storage resources can be called in an emergency state of a power grid, so that the local power supply reliability is improved, and the carbon emission of a traditional peak regulation power plant is reduced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent charging technology, and in particular to an intelligent charging scheduling method and system thereof. Background Art

[0002] With the increasing adoption of new energy vehicles, the randomness and concentration of charging loads pose challenges to grid stability. Existing scheduling methods, often based on fixed electricity price periods or simple priority rankings, fail to dynamically respond to real-time grid load fluctuations and user behavior changes. They focus solely on peak load shaving and valley filling or minimizing user costs, neglecting the collaboration of multiple stakeholders. They rely on historical load data and fail to incorporate multi-dimensional information such as user travel, weather, and traffic. Existing technologies propose charging scheduling based on time-of-use electricity prices, but these methods fail to account for users' urgent charging needs. Furthermore, they are computationally complex and lack real-time performance. Therefore, an intelligent scheduling method with dynamic adjustment and high prediction accuracy is urgently needed. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design an intelligent charging scheduling method and system.

[0004] To achieve the above object, the technical solution of the present invention is that, further, in the above-mentioned intelligent charging scheduling method, the intelligent charging scheduling method includes the following steps: Collecting real-time multi-source data sets, the real-time multi-source data sets including at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends; Establishing an LSTM charging demand prediction model based on an LSTM neural network, inputting the real-time multi-source data set into the LSTM charging demand prediction model for prediction, and obtaining the charging demand and grid dispatchable margin; defining a dynamic priority weight function according to the charging demand and the grid dispatchable margin, and generating a priority list; A multi-objective hierarchical optimization algorithm is established using MIP mixed integer programming and NSGA-II genetic algorithm, and a charging scheduling solution is solved hierarchically based on the multi-objective hierarchical optimization algorithm and with the priority list as a constraint; Charging load scheduling is performed based on the charging scheduling plan. If a power grid failure and user change demand are detected, the multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling plan.

[0005] Furthermore, in the above-mentioned intelligent charging scheduling method, the real-time multi-source data set is collected, and the real-time multi-source data set includes at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends, including: Collect user behavior data through on-board terminal devices and built-in sensors of charging piles, including at least the user's charging time preference, charging duration, charging amount, charging start and end time, vehicle type, and user charging reservation information; Use the power monitoring system to collect real-time grid status data, including at least the real-time voltage, current, frequency, power flow distribution, transformer load factor, and line transmission capacity of the grid; Acquire environmental data based on sensors, including at least ambient temperature, humidity, wind speed, and light intensity; Obtain electricity price fluctuation trends in the electricity price publishing system, including at least real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future.

[0006] Furthermore, in the above-mentioned intelligent charging scheduling method, the LSTM charging demand prediction model is established based on the LSTM neural network, including: The LSTM charging demand prediction model uses three LSTM layers, with 64 neurons in each layer. The output layer uses a linear activation function to output continuous charging demand and grid dispatchable margin prediction values. The optimizer of the LSTM charging demand prediction model is set to Adam optimizer, the initial value of the learning rate is 0.001, and the number of iterations is set to 300.

[0007] Furthermore, in the above-mentioned intelligent charging scheduling method, inputting the real-time multi-source data set into the LSTM charging demand prediction model for prediction to obtain charging demand and grid dispatchable margin includes: The missing values of the real-time multi-source dataset are processed using the mean difference method, and outliers are identified and eliminated using the 3σ principle to obtain a complete real-time multi-source dataset; Normalizing the complete real-time multi-source dataset to the interval [0, 1] based on a minimum-maximum normalization method to obtain a normalized real-time multi-source dataset; The normalized real-time multi-source dataset is divided into a 70% training dataset, a 15% validation dataset, and a 15% test dataset; The training data set is input into the LSTM charging demand prediction model to calculate the predicted value through forward propagation. The error between the predicted value and the true value is calculated using the loss function. The model parameters are updated through the backpropagation algorithm to minimize the loss function and obtain the charging demand and grid dispatchable margin.

[0008] Furthermore, in the above-mentioned intelligent charging scheduling method, defining a dynamic priority weight function according to the charging demand and the grid dispatchable margin to generate a priority list includes: defining a dynamic priority weight function according to the charging demand and the grid dispatchable margin, wherein the priority weight includes user-side urgency, grid-side sensitivity, and charging facility status; The user-side urgency level includes the remaining battery level, the next trip distance, and the charging deadline. The remaining battery level is quantified as follows: if SOC ≤ 20%, the urgency level is +50%. For every 5% decrease in SOC, the score increases linearly by 10%. For each load to be charged, the charging demand and the grid dispatchable margin are substituted into the dynamic priority weight function to calculate the weight value; all loads to be charged are sorted from large to small according to the calculated weight values to generate a priority list.

[0009] Furthermore, in the above-mentioned intelligent charging scheduling method, the multi-objective hierarchical optimization algorithm is established by using MIP mixed integer programming and NSGA-II genetic algorithm. Based on the multi-objective hierarchical optimization algorithm, the charging scheduling scheme is solved hierarchically with the priority list as a constraint, including: The MIP mixed integer programming is used to process discrete variables and continuous variables in charging scheduling, establish a mathematical model, and solve the local optimal solution under the conditions of satisfying grid constraints and user constraints; The local optimal solution obtained by MIP mixed integer programming is used as the initial population. The Pareto frontier solution set is searched in the multi-objective space through selection, crossover and mutation genetic operations to obtain the optimal charging scheduling scheme that meets multiple objectives.

[0010] Furthermore, in the above-mentioned intelligent charging scheduling method, the charging load scheduling is performed based on the charging scheduling plan. If a power grid failure and a user change demand are detected, the target charging scheduling plan is solved using the multi-objective hierarchical optimization algorithm, including: Through the real-time grid status monitoring system, the fault diagnosis algorithm is used to detect grid faults in real time, including short circuit faults, open circuit faults and voltage sags; When a grid fault is detected, the charging of non-critical loads is immediately suspended, and a multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling plan based on the fault type and severity.

[0011] Furthermore, in an intelligent charging scheduling system, the intelligent charging scheduling system includes the following modules: A multi-source data acquisition module is used to collect real-time multi-source data sets, wherein the real-time multi-source data sets include at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends; A demand scheduling prediction module is used to establish an LSTM charging demand prediction model based on an LSTM neural network, input the real-time multi-source data set into the LSTM charging demand prediction model to perform prediction, and obtain charging demand and grid dispatchable margin; A priority judgment module is used to define a dynamic priority weight function according to the charging demand and the grid dispatchable margin to generate a priority list; a scheduling scheme generation module, configured to establish a multi-objective hierarchical optimization algorithm using MIP mixed integer programming and NSGA-II genetic algorithm, and to solve a charging scheduling scheme hierarchically based on the multi-objective hierarchical optimization algorithm and with the priority list as a constraint; The scheduling scheme adjustment module is used to perform charging load scheduling based on the charging scheduling scheme. If a power grid failure and user change demand are detected, the multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling scheme.

[0012] Furthermore, in an intelligent charging scheduling system, the multi-source data acquisition module includes the following submodules: The behavior data acquisition submodule is used to collect user behavior data through the vehicle terminal device and the built-in sensors of the charging pile, including at least the user's charging time preference, charging duration, charging amount, charging start and end time, vehicle type and user charging reservation information; The state data acquisition submodule is used to collect real-time state data of the power grid using the power monitoring system, including at least the real-time voltage, current, frequency, power flow distribution, transformer load rate and line transmission capacity of the power grid; Environmental data acquisition submodule, used to obtain environmental data based on sensors, including at least ambient temperature, humidity, wind speed and light intensity; The electricity price fluctuation acquisition submodule is used to obtain the electricity price fluctuation trend in the electricity price publishing system, including at least real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future.

[0013] Furthermore, in an intelligent charging scheduling system, the multi-source data acquisition module includes the following submodules: The layer number setting submodule uses a three-layer LSTM layer for the LSTM charging demand prediction model, with 64 neurons in each layer. The output layer uses a linear activation function to output continuous charging demand and grid dispatchable margin prediction values. The learning rate setting submodule is used to set the optimizer of the LSTM charging demand prediction model to the Adam optimizer, the initial value of the learning rate to 0.001, and the number of iterations to 300.

[0014] The beneficial effects of this approach include collecting real-time multi-source data sets, establishing an LSTM charging demand forecasting model based on an LSTM neural network, and inputting this real-time multi-source data set into the LSTM charging demand forecasting model for prediction, thereby obtaining charging demand and grid dispatch margin. A dynamic priority weighting function is defined based on the charging demand and grid dispatch margin to generate a priority list. A multi-objective hierarchical optimization algorithm is developed using MIP mixed integer programming and the NSGA-II genetic algorithm. Based on this algorithm and the priority list as constraints, a hierarchical charging scheduling solution is solved. A hierarchical response strategy is employed to proactively shift peak loads, reduce regional peak loads, delay transformer expansion investment, and extend equipment life. Combined with V2G reverse power supply capabilities, electric vehicle energy storage resources can be utilized in grid emergencies, improving local power supply reliability and reducing carbon emissions from traditional peak-shaving power plants. By predicting user behavior and preemptively responding to sudden demand, the satisfaction rate of users' emergency charging needs is increased, and average charging wait times are shortened. This improves load balancing across charging facilities, optimizes peak utilization of fast-charging piles, and reduces equipment idle losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 This is a schematic diagram of a first embodiment of an intelligent charging scheduling method according to an embodiment of the present invention; Figure 2 Schematic diagram of a second embodiment of an intelligent charging scheduling method according to an embodiment of the present invention; Figure 3 Schematic diagram of a first embodiment of an intelligent charging scheduling system in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, a smart charging scheduling method includes the following steps: Step 101: Collect a real-time multi-source data set, which includes at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends; Specifically, in this embodiment, user behavior data is collected through on-board terminal equipment and built-in sensors of charging piles, including at least the user's charging time preference, charging duration, charging amount, charging start and end time, vehicle type and user charging reservation information; the power grid real-time status data is collected using the power monitoring system, including at least the real-time voltage, current, frequency, power flow distribution, transformer load rate and line transmission capacity of the power grid; environmental data is obtained based on sensors, including at least ambient temperature, humidity, wind speed and light intensity; and electricity price fluctuation trends in the electricity price publishing system are obtained, including at least real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future.

[0020] User behavior data collection: Data types: including user charging time preferences (charging after daily commuting, charging during off-peak hours at night, etc.), historical charging records (charging duration, charging amount, charging start and end time), vehicle type (different models have different battery capacities and charging power), user charging reservation information, etc.

[0021] Collection method: Data is transmitted to the data center in real time using wireless communication technology (4G, 5G, Bluetooth) through on-board terminal devices, built-in sensors in charging piles, mobile phone charging apps, etc.

[0022] Collection frequency: Based on the dynamic changes in user behavior, data is collected once every minute to ensure the timeliness and accuracy of the data.

[0023] Real-time power grid status collection: Data type: covers the real-time voltage, current, frequency, power flow distribution, transformer load rate, line transmission capacity, etc. of the power grid.

[0024] Collection method: Relying on smart meters, power monitoring and control systems (SCADA), phasor measurement units (PMUs) and other equipment, optical fiber communication or dedicated power communication networks are used to transmit data quickly and stably to the dispatching center.

[0025] Collection frequency: To meet the high-precision requirements of real-time grid dispatching, the data is collected once per second.

[0026] Environmental data collection: Data types: Mainly include ambient temperature, humidity, wind speed, and light intensity. Ambient temperature has a significant impact on battery charging efficiency and lifespan. Light intensity and wind speed can be used to assess the contribution of renewable energy generation (solar and wind) to the power grid.

[0027] Collection method: Deploy a meteorological sensor network, including temperature sensors, humidity sensors, wind speed sensors, light sensors, etc., and upload the data to the data processing platform through the Internet of Things communication protocol (LoRa, Zigbee).

[0028] Collection frequency: once every 10 minutes. In extreme weather conditions, the high-frequency collection mode can be automatically triggered to collect data once every 1 minute.

[0029] Electricity price fluctuation trend collection: Data types: including real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future (which can be obtained from the electricity market trading platform).

[0030] Collection method: Through interfacing with the power company's electricity price publishing system, electricity price data is obtained regularly using the API interface and stored in the local database.

[0031] Collection frequency: Real-time electricity prices are collected every 15 minutes, and time-of-use electricity prices and electricity price forecast data are updated once a day.

[0032] Step 102: Establish an LSTM charging demand prediction model based on an LSTM neural network, input the real-time multi-source data set into the LSTM charging demand prediction model for prediction, and obtain the charging demand and grid dispatchable margin; Specifically, in this embodiment, the LSTM charging demand forecasting model uses three LSTM layers, each with 64 neurons. The output layer uses a linear activation function to output continuous charging demand and grid dispatch margin prediction values. The Adam optimizer is set as the optimizer for the LSTM charging demand forecasting model, with an initial learning rate of 0.001 and 300 iterations. Missing values in the real-time multi-source dataset are processed using mean interpolation, and outliers are identified and removed using the 3σ principle to obtain a complete real-time multi-source dataset. The complete real-time multi-source dataset is normalized to the interval [0, 1] using the min-max normalization method to obtain a normalized real-time multi-source dataset. The normalized real-time multi-source dataset is divided into a 70% training dataset, a 15% validation dataset, and a 15% test dataset. The training dataset is input into the LSTM charging demand forecasting model, and the predicted value is calculated through forward propagation. The error between the predicted value and the true value is calculated using a loss function. The model parameters are updated through a backpropagation algorithm to minimize the loss function, thereby obtaining the charging demand and grid dispatch margin.

[0033] Data preprocessing: Data cleaning: missing values are processed for the collected multi-source data, and missing data are filled using mean interpolation, median interpolation, or model-based predictive interpolation methods; outliers are identified and eliminated, for example, using the 3σ principle (data outside the mean ± 3 times the standard deviation range is considered abnormal) for outlier detection and processing. Data normalization: normalize data of different dimensions to the range of [0,1] or [-1,1]. Data partitioning: divide the processed data into a training set (accounting for 70%), a validation set (accounting for 15%), and a test set (accounting for 15%). LSTM neural network model construction: Network structure design: Determine the number of layers, neurons, and activation function for the LSTM neural network. Typically, two to three LSTM layers are used. The number of neurons in each layer is adjusted based on data characteristics and computing resources, with examples being 128 or 64. The output layer uses a linear activation function to output continuous charging demand and grid dispatch margin predictions.

[0034] Parameter setting: Select an appropriate optimizer (Adam optimizer), set parameters such as learning rate (the initial value can be set to 0.001), number of iterations (determined according to the training effect, generally 100-500 times), batch size (32, 64), etc.

[0035] Model training and optimization: Training process: The training set data is input into the LSTM neural network model, the predicted value is calculated through forward propagation, the error between the predicted value and the true value is calculated using the loss function (mean square error MSE), and then the model parameters are updated through the backpropagation algorithm to minimize the loss function.

[0036] Model evaluation and adjustment: The model during training is evaluated on the validation set. Model parameters and structure are adjusted based on evaluation metrics (root mean square error (RMSE) and mean absolute error (MAE)) to prevent overfitting and underfitting. When the model achieves optimal performance on the validation set, a final test is conducted on the test set to obtain a stable and reliable LSTM charging demand prediction model.

[0037] Prediction process: The multi-source data set collected in real time and pre-processed is input into the trained LSTM charging demand prediction model, and the model outputs the charging demand forecast value and the grid dispatchable margin forecast value for a period of time in the future (1 hour, 2 hours).

[0038] Step 103: Define a dynamic priority weight function based on charging demand and grid dispatchable margin to generate a priority list; Specifically, in this embodiment, a dynamic priority weight function is defined based on the charging demand and the grid dispatchable margin, where the priority weight includes the user-side urgency, the grid-side sensitivity and the charging facility status; the user-side urgency includes the remaining power, the next travel distance and the charging deadline, and the remaining power is quantified as SOC ≤ 20%, urgency + 50%; for every 5% decrease in SOC, the score increases linearly by 10%; for each load to be charged, the charging demand and the grid dispatchable margin are substituted into the dynamic priority weight function to calculate the weight value; all loads to be charged are sorted from large to small according to the calculated weight values to generate a priority list.

[0039] The quantification method for the next trip distance is distance / full-charge range ≥ 80% → urgency + 30%; the quantification method for the charging deadline (user-set) is remaining time < 2 hours → urgency + 20%, and the score increases by 10% for every 30 minutes of reduction.

[0040] Grid-side sensitivity includes regional transformer load factor, node voltage deviation, and grid frequency fluctuation. Regional transformer load factor is quantified as follows: load factor ≥ 90% → sensitivity weight +50%; for every 10% increase in load factor, the score increases by 20%. Node voltage deviation is quantified as follows: voltage deviation ±5% of rated value → sensitivity +30%, with an excess of ±7% triggering emergency load reduction. Grid frequency fluctuation is quantified as frequency deviation > 0.2 Hz → sensitivity weight increased to the highest priority.

[0041] Charging facility status includes charging pile availability, current queue length, and charging pile type. Charging pile availability is quantified as follows: when the number of faulty piles exceeds 30%, the load factor score increases by 40%, triggering charging task migration. The current queue length is quantified as follows: when there are more than five vehicles in the queue, the load factor score increases by 20%, initiating dynamic diversion (directing vehicles to nearby charging stations). Charging pile type is quantified as follows: a fast charging pile (120kW) is assigned a load factor weight three times that of a slow charging pile (7kW) (due to its greater impact on the power grid).

[0042] The objectives of multi-objective optimization also include: multi-objective definition, Goal 1: Maximize user satisfaction: Considering factors such as the user's charging time preference and the degree to which their charging needs are met, the deviation between the user's actual charging time and the expected charging time is minimized as a measurement indicator.

[0043] Goal 2: Minimize grid operation costs: This includes active power loss, reactive power compensation costs, peak-valley adjustment costs, etc., with the goal of minimizing the total cost during grid operation.

[0044] Goal 3: Minimize charging costs: According to the fluctuation trend of electricity prices, reasonably arrange charging time to minimize charging costs.

[0045] Combination of MIP mixed integer programming and NSGA-II genetic algorithm: MIP mixed integer programming: used to process discrete variables (charging station on / off status, charging time selection, etc.) and continuous variables (charging power) in charging scheduling, establish mathematical models, and solve local optimal solutions while satisfying grid constraints (power balance constraints, voltage constraints, line capacity constraints, etc.) and user constraints (charging power limits, charging time window constraints, etc.).

[0046] NSGA-II genetic algorithm: uses the local optimal solution obtained by MIP mixed integer programming as the initial population, and searches for the global optimal solution or Pareto frontier solution set in the multi-objective space through genetic operations such as selection, crossover, and mutation to find the optimal charging scheduling solution that simultaneously meets multiple objectives.

[0047] Hierarchical solution process: The first layer: Priority list screening: Based on the generated priority list, the loads to be charged with higher priority are screened out, and the charging scheduling plan is solved for these loads first to ensure that the charging needs of important users are met.

[0048] Second Layer: Multi-Objective Hierarchical Optimization: Building on the first-layer screening, a multi-objective hierarchical optimization algorithm is established, with the goals of maximizing user satisfaction, minimizing grid operating costs, and minimizing charging costs. This algorithm combines the constraints of the grid and users to solve a charging scheduling solution. During the solution process, the user satisfaction objective is first optimized to obtain a set of solutions. Then, while maintaining user satisfaction, the grid operating cost objective is optimized. Finally, while maintaining the first two objectives, the charging cost objective is optimized, gradually obtaining the optimal overall charging scheduling solution.

[0049] Step 104: Establish a multi-objective hierarchical optimization algorithm using MIP mixed integer programming and NSGA-II genetic algorithm. Based on the multi-objective hierarchical optimization algorithm, use the priority list as a constraint to solve the charging scheduling plan in a hierarchical manner. Specifically, in this embodiment, MIP mixed integer programming is used to process discrete variables and continuous variables in charging scheduling, establish a mathematical model, and solve the local optimal solution while satisfying grid constraints and user constraints; the local optimal solution obtained by MIP mixed integer programming is used as the initial population, and the Pareto frontier solution set is searched in the multi-objective space through selection, crossover and mutation genetic operations to obtain the optimal charging scheduling solution that meets multiple objectives.

[0050] Weight function design: Analysis of influencing factors: Consider factors such as the urgency of charging demand (urgent charging needs of users about to depart have a higher weight), the size of the grid's dispatchable margin (when the margin is small, priority is given to charging needs with less impact on the grid), and electricity price costs (during periods of high electricity prices, low-priority users postpone charging).

[0051] Step 105: Perform charging load scheduling based on the charging scheduling plan. If a grid failure or user change demand is detected, a multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling plan.

[0052] Normal charging load scheduling: The resulting charging schedule is sent to the charging pile control system, which then charges electric vehicles and other loads in an orderly manner according to the schedule's time and power. During the charging process, the charging status and grid operation status are monitored in real time, and charging data (actual charging power, charge volume, charging time, etc.) is recorded for subsequent analysis and optimization.

[0053] Grid fault response: Fault detection: Through the real-time status monitoring system of the power grid, the fault diagnosis algorithm (neural network-based fault diagnosis method, traveling wave method, etc.) is used to detect power grid faults in real time, including short circuit faults, open circuit faults, voltage sags, etc.

[0054] Adjustment of the dispatching plan: Once a grid fault is detected, charging of non-critical loads is immediately suspended. Based on the type and severity of the fault, the multi-objective hierarchical optimization algorithm is reused to ensure the safe and stable operation of the grid and power supply to important users as the primary goal. The charging dispatching plan is adjusted to prioritize the power demand of important loads such as hospitals and traffic lights, while minimizing the impact on user charging.

[0055] Response to user change requirements: Demand change detection: Receive real-time information on changes in user charging needs through channels such as the user charging app and charging pile operation interface, and advance, postpone, or cancel charging time.

[0056] Re-optimization of the dispatch plan: When receiving user change requirements, the multi-objective hierarchical optimization algorithm is used to re-solve the charging dispatch plan based on the new demand information, combined with the current grid operation status and the charging dispatch of other users. Under the premise of meeting the grid constraints and other user needs, the user's change requirements are met as much as possible to improve user satisfaction.

[0057] Specifically, in this embodiment, a real-time grid status monitoring system is used to detect grid faults in real time using a fault diagnosis algorithm, including short circuit faults, open circuit faults, and voltage sags. When a grid fault is detected, charging of non-critical loads is immediately suspended, and a multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling plan based on the fault type and severity.

[0058] The proposed model collects real-time multi-source data sets and establishes an LSTM charging demand forecasting model based on an LSTM neural network. This data set is then fed into the LSTM charging demand forecasting model to generate predictions, deriving charging demand and grid dispatch margin. A dynamic priority weighting function is defined based on charging demand and grid dispatch margin to generate a priority list. A multi-objective hierarchical optimization algorithm is developed using MIP mixed integer programming and the NSGA-II genetic algorithm. Based on this algorithm and the priority list as constraints, a hierarchical charging scheduling solution is generated. A hierarchical response strategy is employed to proactively shift peak loads, reduce regional peak loads, delay transformer expansion investments, and extend equipment life. Combined with V2G reverse power supply capabilities, electric vehicle energy storage resources can be deployed during grid emergencies, improving local power supply reliability and reducing carbon emissions from traditional peak-shaving power plants. By predicting user behavior and preemptively responding to sudden demand, the satisfaction rate of users' emergency charging needs is increased, and average charging wait times are shortened. This improves load balancing across charging facilities, optimizes peak utilization of fast-charging stations, and reduces equipment idle losses.

[0059] See also Figure 2 In an intelligent charging scheduling method, collecting a real-time multi-source data set, which includes at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends, includes the following steps: Step 201: Collect user behavior data through the vehicle terminal device and the built-in sensors of the charging pile, including at least the user's charging time preference, charging duration, charging amount, charging start and end time, vehicle type, and user charging reservation information; Step 202: Using a power monitoring system to collect real-time grid status data, including at least the real-time voltage, current, frequency, power flow distribution, transformer load factor, and line transmission capacity of the grid; Step 203: Acquire environmental data based on sensors, including at least ambient temperature, humidity, wind speed, and light intensity; Step 204: Obtain electricity price fluctuation trends in the electricity price publishing system, including at least real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future.

[0060] The above is an introduction to an embodiment of an intelligent charging scheduling method of the present invention. Figure 3In an intelligent charging scheduling system, the intelligent charging scheduling system includes the following modules: A multi-source data acquisition module is used to collect real-time multi-source data sets, which include at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends; The demand scheduling forecasting module is used to establish an LSTM charging demand forecasting model based on the LSTM neural network. The real-time multi-source data set is input into the LSTM charging demand forecasting model to predict the charging demand and the grid dispatchable margin. Priority judgment module, used to define dynamic priority weight function according to charging demand and grid dispatchable margin, and generate priority list; The scheduling scheme generation module is used to establish a multi-objective hierarchical optimization algorithm using MIP mixed integer programming and NSGA-II genetic algorithm. Based on the multi-objective hierarchical optimization algorithm, the charging scheduling scheme is solved hierarchically with the priority list as the constraint; The scheduling scheme adjustment module is used to schedule charging loads based on the charging scheduling scheme. If a grid failure or user change demand is detected, a multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling scheme.

[0061] Specifically, in this embodiment, the application of NSGA-II genetic algorithm in charging scheduling is explained NSGA-II (Non-Dominated Sorting Genetic Algorithm II) is a highly efficient algorithm specifically designed to solve multi-objective optimization problems. In the charging scheduling scenario described in this paper, the three objectives of grid peak shifting, minimizing user costs, and maximizing equipment utilization often conflict with each other (for example, reducing grid peaks may require delaying charging for some users, increasing their waiting costs). The core value of NSGA-II lies in simultaneously exploring the optimal balance between multiple objectives, rather than simply compromising or prioritizing a single objective. Its specific implementation logic is as follows: Core idea: Finding the “Pareto optimal” solution set Assume that there is a set of charging scheduling schemes: if a scheme is not inferior to other schemes in terms of both "reducing grid peak" and "reducing user costs", and is better in at least one of the two objectives, then the scheme is "Pareto optimal". The goal of NSGA-II is to find all such schemes that cannot be further optimized and form a diverse candidate solution ( Figure 1 shown).

[0062] Pareto frontier in multi-objective optimization (horizontal and vertical axes represent different optimization objectives) 1. Role in charging scheduling: The algorithm generates hundreds of possible charging plans (vehicle A charges at 18:00 and vehicle B charges at 22:00), from which it selects the "high-quality solution group" that can alleviate the pressure on the power grid without significantly increasing user costs, providing decision makers with flexible choices.

[0063] 2. Key steps and charging scheduling scenario mapping Although no formulas are involved, the process can be described in natural language: Initialize the population: A batch of charging scheduling plans are randomly generated (charging time periods are randomly assigned), and each plan must meet the basic constraints (the vehicle must eventually be fully charged).

[0064] Non-dominated sorting: Ranking of options by merit: Level 1: The solution that is not surpassed by other solutions in all objectives (i.e., the Pareto optimal solution); Level 2: Solutions that are only surpassed by Level 1 solutions; And so on.

[0065] Scenario mapping: If solution X can reduce both grid load variance and user total cost compared to solution Y, then X is ranked higher than Y.

[0066] Crowding calculation: Among the solutions at the same level, the solution with “sparse” target space distribution is prioritized to avoid the algorithm converging to the local optimum.

[0067] Scenario mapping: Ensure that candidate solutions include both "aggressive grid peak reduction" solutions and "extremely low user cost" solutions, rather than focusing entirely on intermediate states.

[0068] Selection and evolution: By simulating the "crossover" and "mutation" in biological evolution, a new generation of solutions is generated: Crossover: Mix the charging time periods of the two best plans (the first half comes from plan A, and the second half comes from plan B). Mutation: Randomly adjust the charging time or power of a vehicle to explore new possibilities.

[0069] Iteration convergence: Repeat the above steps until enough Pareto optimal solutions are found. Finally, the system or user selects the final solution based on real-time preferences (for example, when the grid is overloaded, the solution with the best peak shaving effect is prioritized).

[0070] 3. Unique advantages in charging scheduling Multi-target collaboration: Traditional methods might weight multiple objectives into a single composite metric (e.g., 0.7 × grid load + 0.3 × user cost), but the choice of weights is subjective. NSGA-II directly presents multiple possible trade-offs, avoiding the influence of human bias.

[0071] Dynamic adaptability: When the grid status changes suddenly (transformer failure) or the user temporarily modifies the itinerary, NSGA-II can quickly recalculate and generate a new Pareto solution set to adapt to real-time changes.

[0072] Fairness in resource allocation: The congestion mechanism is used to ensure the diversity of solutions, prevent certain user groups (vehicles with non-emergency needs) from being sacrificed for a long time, and improve the acceptability of the solution.

[0073] 4. Example Assuming that there are 10 electric vehicles in a community that need to be charged, NSGA-II may give two typical solutions: Solution A: 5 cars are charged during the night off-peak period, which minimizes grid load variance, but users of 3 cars need to wait 4 more hours. Solution B: The eight cars are charged in the evening when electricity prices are higher. This results in the lowest total cost for users, but the peak power consumption of the grid increases by 10%.

[0074] The system can automatically select solution A or solution B based on the current grid load rate (whether it is close to the transformer upper limit), rather than forcing the use of a single strategy.

[0075] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent charging scheduling method, characterized in that: The intelligent charging scheduling method comprises the following steps: Collecting real-time multi-source data sets, the real-time multi-source data sets including at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends; Establishing an LSTM charging demand prediction model based on an LSTM neural network, inputting the real-time multi-source data set into the LSTM charging demand prediction model for prediction, and obtaining the charging demand and grid dispatchable margin; defining a dynamic priority weight function according to the charging demand and the grid dispatchable margin, and generating a priority list; A multi-objective hierarchical optimization algorithm is established using MIP mixed integer programming and NSGA-II genetic algorithm, and a charging scheduling solution is solved hierarchically based on the multi-objective hierarchical optimization algorithm and with the priority list as a constraint; Charging load scheduling is performed based on the charging scheduling plan. If a power grid failure and user change demand are detected, the multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling plan.

2. The intelligent charging scheduling method according to claim 1, characterized in that: The real-time multi-source data set collected includes at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends, including: Collect user behavior data through on-board terminal devices and built-in sensors of charging piles, including at least the user's charging time preference, charging duration, charging amount, charging start and end time, vehicle type, and user charging reservation information; Use the power monitoring system to collect real-time grid status data, including at least the real-time voltage, current, frequency, power flow distribution, transformer load factor, and line transmission capacity of the grid; Acquire environmental data based on sensors, including at least ambient temperature, humidity, wind speed, and light intensity; Obtain electricity price fluctuation trends in the electricity price publishing system, including at least real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future.

3. The intelligent charging scheduling method according to claim 1, characterized in that: The LSTM charging demand prediction model is established based on the LSTM neural network, including: The LSTM charging demand prediction model uses three LSTM layers, with 64 neurons in each layer. The output layer uses a linear activation function to output continuous charging demand and grid dispatchable margin prediction values. The optimizer of the LSTM charging demand prediction model is set to Adam optimizer, the initial value of the learning rate is 0.001, and the number of iterations is set to 300.

4. The intelligent charging scheduling method according to claim 1, characterized in that: Inputting the real-time multi-source data set into the LSTM charging demand prediction model to perform prediction to obtain charging demand and grid dispatchable margin includes: The missing values of the real-time multi-source dataset are processed using the mean difference method, and outliers are identified and eliminated using the 3σ principle to obtain a complete real-time multi-source dataset; Normalizing the complete real-time multi-source dataset to the interval [0, 1] based on a minimum-maximum normalization method to obtain a normalized real-time multi-source dataset; The normalized real-time multi-source dataset is divided into a 70% training dataset, a 15% validation dataset, and a 15% test dataset; The training data set is input into the LSTM charging demand prediction model to calculate the predicted value through forward propagation. The error between the predicted value and the true value is calculated using the loss function. The model parameters are updated through the backpropagation algorithm to minimize the loss function and obtain the charging demand and grid dispatchable margin.

5. The intelligent charging scheduling method according to claim 1, characterized in that: Defining a dynamic priority weight function according to the charging demand and the grid dispatchable margin to generate a priority list includes: defining a dynamic priority weight function according to the charging demand and the grid dispatchable margin, wherein the priority weight includes user-side urgency, grid-side sensitivity, and charging facility status; The user-side urgency level includes the remaining battery level, the next trip distance, and the charging deadline. The remaining battery level is quantified as follows: if SOC ≤ 20%, the urgency level is +50%. For every 5% decrease in SOC, the score increases linearly by 10%. For each load to be charged, the charging demand and the grid dispatchable margin are substituted into the dynamic priority weight function to calculate the weight value; all loads to be charged are sorted from large to small according to the calculated weight values to generate a priority list.

6. The intelligent charging scheduling method according to claim 1, characterized in that: The multi-objective hierarchical optimization algorithm is established by using MIP mixed integer programming and NSGA-II genetic algorithm. Based on the multi-objective hierarchical optimization algorithm, the charging scheduling scheme is solved hierarchically with the priority list as a constraint, including: The MIP mixed integer programming is used to process discrete variables and continuous variables in charging scheduling, establish a mathematical model, and solve the local optimal solution under the conditions of satisfying grid constraints and user constraints; The local optimal solution obtained by MIP mixed integer programming is used as the initial population. The Pareto frontier solution set is searched in the multi-objective space through selection, crossover and mutation genetic operations to obtain the optimal charging scheduling scheme that meets multiple objectives.

7. The intelligent charging scheduling method according to claim 1, characterized in that: The charging load scheduling is performed based on the charging scheduling plan. If a power grid failure and a user change demand are detected, the target charging scheduling plan is solved by using the multi-objective hierarchical optimization algorithm, including: Through the real-time grid status monitoring system, the fault diagnosis algorithm is used to detect grid faults in real time, including short circuit faults, open circuit faults and voltage sags; When a grid fault is detected, the charging of non-critical loads is immediately suspended, and a multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling plan based on the fault type and severity.

8. An intelligent charging scheduling system, characterized in that: The intelligent charging scheduling system includes the following modules: A multi-source data acquisition module is used to collect real-time multi-source data sets, wherein the real-time multi-source data sets include at least user behavior data, real-time grid status data, environmental data, and electricity price fluctuation trends; A demand scheduling prediction module is used to establish an LSTM charging demand prediction model based on an LSTM neural network, input the real-time multi-source data set into the LSTM charging demand prediction model to perform prediction, and obtain charging demand and grid dispatchable margin; A priority judgment module is used to define a dynamic priority weight function according to the charging demand and the grid dispatchable margin to generate a priority list; a scheduling scheme generation module, configured to establish a multi-objective hierarchical optimization algorithm using MIP mixed integer programming and NSGA-II genetic algorithm, and to solve a charging scheduling scheme hierarchically based on the multi-objective hierarchical optimization algorithm and with the priority list as a constraint; The scheduling scheme adjustment module is used to perform charging load scheduling based on the charging scheduling scheme. If a power grid failure and user change demand are detected, the multi-objective hierarchical optimization algorithm is used to solve the target charging scheduling scheme.

9. The intelligent charging scheduling system according to claim 8, characterized in that: The multi-source data acquisition module includes the following submodules: The behavior data acquisition submodule is used to collect user behavior data through the vehicle terminal device and the built-in sensors of the charging pile, including at least the user's charging time preference, charging duration, charging amount, charging start and end time, vehicle type and user charging reservation information; The state data acquisition submodule is used to collect real-time state data of the power grid using the power monitoring system, including at least the real-time voltage, current, frequency, power flow distribution, transformer load rate and line transmission capacity of the power grid; Environmental data acquisition submodule, used to obtain environmental data based on sensors, including at least ambient temperature, humidity, wind speed and light intensity; The electricity price fluctuation acquisition submodule is used to obtain the electricity price fluctuation trend in the electricity price publishing system, including at least real-time electricity prices, time-of-use electricity prices, peak and valley electricity price period divisions, and electricity price forecast data for a period of time in the future.

10. The intelligent charging scheduling system according to claim 8, characterized in that: The multi-source data acquisition module includes the following submodules: The layer number setting submodule uses a three-layer LSTM layer for the LSTM charging demand prediction model, with 64 neurons in each layer. The output layer uses a linear activation function to output continuous charging demand and grid dispatchable margin prediction values. The learning rate setting submodule is used to set the optimizer of the LSTM charging demand prediction model to the Adam optimizer, the initial value of the learning rate to 0.001, and the number of iterations to 300.

Images