AI-based dynamic charging power adjustment and intelligent allocation system

Abstract

This invention provides an AI-based dynamic adjustment and intelligent allocation system for charging power, relating to the field of charging power technology. The system includes: collecting charging data and preprocessing the data; using an N-BEATS model to analyze the time series relationships in the data to obtain the charging demand distribution at time t; using a graph neural network to predict the regional power grid load and analyze the spatial coupling relationship between charging piles to obtain the upper limit of the available power of the power grid at future time t; performing an initial allocation of charging power using a fuzzy control algorithm; and recalculating the power allocation scheme using the fuzzy control algorithm to update the prediction results of the N-BEATS model, resulting in a clear charging power allocation. This invention combines the N-BEATS model with a fuzzy control algorithm to generate charging power allocation, ensuring the scientific and rational nature of the power allocation and improving the utilization rate of charging equipment.

Description

Technical Field

[0001] This invention relates to the field of charging power technology, and in particular to an AI-based dynamic adjustment and intelligent distribution system for charging power. Background Technology

[0002] In the current charging landscape, the mismatch between the number of charging stations and the number of electric vehicles is becoming increasingly apparent. Especially during peak hours, traditional charging power allocation methods often have numerous shortcomings. They struggle to effectively process multi-source, heterogeneous charging data, accurately extract key features, capture long-distance dependencies in charging demand, and provide inaccurate predictions of future charging needs. Furthermore, they cannot quantify the spatial coupling relationships between charging stations, accurately predict regional grid load, lack a comprehensive consideration of supply and demand, and employ rigid allocation strategies that fail to adapt to dynamically changing charging scenarios and grid conditions. This makes it difficult to adjust power allocation schemes in a timely manner based on actual conditions, resulting in low charging efficiency, unbalanced grid load, and an inability to effectively guarantee battery life.

[0003] With the rapid development of artificial intelligence technology, its application in various fields is becoming increasingly widespread. Applying AI technology to the regulation and distribution of charging power provides a new approach to solving the aforementioned problems. Summary of the Invention

[0004] This invention provides an AI-based dynamic adjustment and intelligent allocation system for charging power, which solves the problem in the prior art that it is difficult to adjust the power allocation scheme in a timely manner according to the actual situation, resulting in low charging efficiency and unbalanced grid load.

[0005] On one hand, the present invention provides an AI-based dynamic adjustment and intelligent allocation system for charging power, comprising:

[0006] The data acquisition module is used to collect charging data, including real-time charging data and historical charging data.

[0007] The data preprocessing module is used to preprocess the charging data to obtain preprocessed charging data.

[0008] The electricity consumption forecasting module is used to analyze the long-distance dependencies of time series in preprocessed charging data using the N-BEATS model, and obtain the charging demand distribution at time t in the future.

[0009] The power cap module includes: a charging pile topology graph unit, which sets each charging pile in the area as a graph node; a spatial association unit, which quantifies the load coupling strength between charging piles based on the physical connection relationship of the power grid containing the graph nodes, and uses edge weights to obtain a graph structure representing spatial association; a load conduction unit, which aggregates features of the graph structure representing spatial association through a graph convolutional network, where each node collects the load features of its neighboring nodes, uses graph convolution operators to achieve message passing, and obtains the load conduction effect between different charging piles; a spatially dependent node unit, which generates spatially dependent nodes containing the spatial dependency relationship of the load conduction effect between different charging piles after multiple convolutions; and a power cap unit, which concatenates the spatial features of the spatially dependent nodes with historical time series, predicts the load value of each charging pile at time t in the future through a time-series convolutional layer, and calculates the dynamic upper limit of the available power in the area by summing the results and combining them with the power grid capacity constraints.

[0010] The power allocation module is used to initially allocate charging power under the constraints of the charging demand distribution at time t in the future and the upper limit of the available power of the power grid at time t in the future, through a fuzzy control algorithm.

[0011] The real-time feedback module uses a fuzzy control algorithm to recalculate the power allocation scheme, update the prediction results of the N-BEATS model, and obtain a clear charging power allocation.

[0012] According to the AI-based dynamic adjustment and intelligent allocation system for charging power provided by the present invention, the data acquisition module includes:

[0013] The vehicle battery data acquisition unit is used to collect information such as current battery level, battery health status, remaining charging capacity, and charging interface type.

[0014] The charging pile data acquisition unit is used to collect rated power, current load, and available power threshold.

[0015] The power grid data acquisition unit is used to collect real-time voltage, real-time current, power factor, regional load rate, and electricity price during specific time periods.

[0016] The user data acquisition unit is used to collect the target charging capacity, expected completion time, and charging priority.

[0017] The environmental data acquisition unit is used to collect air temperature and humidity.

[0018] According to the AI-based dynamic adjustment and intelligent allocation system for charging power provided by the present invention, the data preprocessing module includes:

[0019] The feature extraction unit is used to extract key features from the charging data.

[0020] The data standardization unit is used to standardize the battery's maximum acceptable charging power and available power resources based on historical charging data.

[0021] The data cleaning unit is used to check for and correct any anomalies in the charging data.

[0022] According to the AI-based dynamic adjustment and intelligent allocation system for charging power provided by the present invention, the processing steps of the power consumption prediction module include:

[0023] Map historical charging data in the preprocessed charging data to an input tensor;

[0024] Construct an N-BEATS deep residual network model;

[0025] The ability of the N-BEATS deep residual network model to capture long-sequence dependencies is optimized by using the cyclic loss function, and the intrinsic structure of time series is learned.

[0026] The long-term variation pattern and short-term periodic pattern are extracted by the trend generator and seasonal generator of the N-BEATS deep residual network model, respectively. The long-term variation pattern and short-term periodic pattern are superimposed to obtain the preliminary distribution of charging demand in the future time t.

[0027] Multiple independent N-BEATS model instances are trained, and differentiated training is performed based on the preliminary distribution of charging demand in the future time t. The preliminary distribution of charging demand in the future time t after differentiated training is then fused to obtain a fused instance.

[0028] Dropout integration is used to perform multiple forward propagations of the fused instance to generate multiple prediction samples. The probability density function of the charging demand is generated by statistical analysis of the prediction samples, the confidence interval of the prediction results is calculated, and the prediction uncertainty is quantified.

[0029] By using an online learning mechanism, the N-BEATS model parameters are adjusted based on real-time charging data to obtain the charging demand distribution at time t in the future.

[0030] According to the AI-based dynamic adjustment and intelligent allocation system for charging power provided by the present invention, the power allocation module includes:

[0031] The variable determination unit is used to determine the input and output variables of fuzzy control, formulate fuzzy rules, and build a rule base.

[0032] The power allocation unit is used to perform fuzzy reasoning based on input and output variables and a rule base, infer a preliminary conclusion on power allocation according to the rules in the rule base, and obtain a fuzzy power allocation result.

[0033] The fuzzy defuzzification unit is used to convert the fuzzy power allocation result into the power allocation value of each charging pile using the centroid method.

[0034] According to the AI-based dynamic adjustment and intelligent allocation system for charging power provided by the present invention, the fuzzy resolution unit includes:

[0035] Membership subunits are used to determine fuzzy sets based on the rules of the rule base and the actual input variables.

[0036] The subset parameter calculation subunit is used to calculate the membership degree corresponding to different power values ​​in each fuzzy subset using the membership function.

[0037] The clear power value calculation subunit is used to substitute the power values ​​and membership degrees of each fuzzy subset into the centroid method calculation formula to calculate the clear power allocation value.

[0038] The membership function of the AI-based dynamic adjustment and intelligent allocation system for charging power provided by this invention is expressed as follows:

[0039]

[0040] In the formula, a and b are parameters that determine the range of the fuzzy set, and x is the battery state of charge.

[0041] The AI-based dynamic adjustment and intelligent allocation system for charging power provided by the present invention includes a real-time feedback module comprising:

[0042] The result update unit is used to update the prediction results of the N-BEATS model using the initially allocated charging power.

[0043] The re-prediction unit is used to adjust the computational logic of the N-BEATS model to obtain a new prediction of the charging demand distribution at time t in the future.

[0044] The feedback adjustment unit is used to input the newly predicted charging demand distribution within the next time t into the fuzzy control algorithm again, so as to obtain the power allocation scheme in the next round of calculation.

[0045] The AI-based dynamic adjustment and intelligent allocation system for charging power provided by this invention also includes a real-time monitoring module. This module is used to: collect power allocation-related data from charging piles and display the real-time power allocation curves for each charging pile; monitor the grid load in real time and display a dynamic graph of the grid load; and use color coding to distinguish charging priorities.

[0046] The AI-based dynamic adjustment and intelligent allocation system for charging power provided by this invention predicts the charging demand distribution at time t in the future using an N-BEATS model and predicts the regional power grid load using a graph neural network, thereby obtaining the upper limit of the available power of the power grid at time t in the future. Then, a fuzzy control algorithm is used to redistribute the charging power, achieving the following beneficial effects:

[0047] The system collects historical and real-time charging data, and performs multi-faceted analysis and prediction based on this data, as well as subsequent power allocation and monitoring feedback. The entire process is data-driven, and the generated predictive analysis reports can provide managers with intuitive and valuable references to help them make scientific and reasonable decisions. For example, they can plan resource allocation in advance based on the distribution of charging demand and power limits, and adjust power allocation strategies in a timely manner based on real-time monitoring feedback, thereby ensuring the high-quality operation of charging services.

[0048] The N-BEATS model can deeply analyze the long-distance dependencies hidden in preprocessed charging data, thereby accurately predicting the charging demand distribution at time t in the future, providing an important time-dimensional reference for subsequent power allocation. Graph neural networks excel at capturing spatial structure information in data. By analyzing the spatial coupling relationships between charging piles, they accurately predict the regional power grid load based on preprocessed charging data, and then derive the upper limit of available power in the power grid. This provides crucial spatial dimension support for considering the collaborative operation of multiple charging piles and the overall carrying capacity of the power grid. Fuzzy control algorithms, with their effective handling of fuzzy and uncertain information, can rationally allocate power under the constraints of charging demand distribution and the upper limit of available power in the power grid, which have certain fuzzy characteristics. Whether in the initial allocation or subsequent real-time feedback adjustments, they can effectively balance various complex factors, ensuring the scientific and rational nature of power allocation.

[0049] Through the ingenious integration of various modules, collaborative work is achieved, ultimately resulting in comprehensive and practical integrated functions. From data acquisition and preprocessing to prepare the "raw materials" for model analysis, to electricity consumption forecasting and power limit prediction to provide key "constraints" for power allocation, to real-time feedback to dynamically optimize the allocation scheme, and real-time monitoring to intuitively present the entire process and generate reports, multiple technical means work together to enable the system not only to achieve reasonable allocation of charging power, but also to monitor the system's operating status in real time and predict future trends in advance. This comprehensively improves the intelligence level and operating efficiency of the charging system, meeting various practical needs during the charging process, such as ensuring grid stability, improving the utilization rate of charging equipment, and optimizing the user charging experience. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 This is a block diagram of the AI-based dynamic adjustment and intelligent allocation system for charging power provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0053] The following is combined Figure 1 This invention describes an AI-based dynamic adjustment and intelligent allocation system for charging power.

[0054] Figure 1 This is a block diagram of the AI-based dynamic adjustment and intelligent allocation system for charging power provided in an embodiment of the present invention.

[0055] like Figure 1 As shown in the figure, the AI-based dynamic adjustment and intelligent allocation system for charging power provided in this embodiment of the invention includes: a data acquisition module, a data preprocessing module, a power consumption prediction module, a power upper limit module, a power allocation module, a real-time feedback module, and a real-time monitoring module.

[0056] The data acquisition module is used to collect charging data in real time.

[0057] In a specific charging scenario, such as a charging area in a large parking lot, multiple smart charging stations are deployed. Each charging station is equipped with sensors and a communication module to collect vehicle battery data in real time, including:

[0058] The vehicle battery data acquisition unit is used to collect information such as current battery level, battery health status, remaining charging capacity, and charging interface type.

[0059] The charging pile data acquisition unit is used to collect rated power, current load, and available power threshold.

[0060] The power grid data acquisition unit is used to collect real-time voltage, real-time current, power factor, regional load rate, and electricity price during specific time periods.

[0061] The user data acquisition unit is used to collect the target charging capacity, expected completion time, and charging priority.

[0062] The environmental data acquisition unit is used to collect air temperature and humidity.

[0063] This data is transmitted to the central control system in real time via wireless communication.

[0064] The data preprocessing module is used to preprocess the charging data to obtain preprocessed charging data. The data preprocessing module includes: a feature extraction unit, a data standardization unit, and a data cleaning unit.

[0065] The feature extraction unit is used to extract key features from charging data. In a specific embodiment, given a current battery charge of 30% and a target charge of 80%, the charging demand ratio is constructed as (80%-30%) / 80%=62.5%. Based on a battery health status of 85% and an ambient temperature of 25℃, combined with historical data and a battery characteristic model, the maximum acceptable charging power feature for the battery is constructed as 50kW. Combining the charging pile's output power range of 0-60kW and a grid load rate of 70%, the available power resource feature is constructed as 60kW*(1-70%)=18kW. Peak hour features are constructed based on the current time.

[0066] The data standardization unit is used to standardize the battery's acceptable maximum charging power and available power resources based on historical charging data. For example, using minimum-maximum standardization, the charging demand ratio is mapped to the [0,1] interval. Assuming the historical maximum charging demand ratio is 100%, the standardized value is 62.5% / 100% = 0.625. Standardizing the battery's acceptable maximum charging power and available power resources according to their historical range ensures that all characteristics are on the same order of magnitude.

[0067] The data cleaning unit is used to check the collected data for anomalies. If a vehicle's battery voltage data is found to be 300V, it is determined to be an outlier and corrected using interpolation or other reasonable methods to ensure the accuracy of the data.

[0068] The electricity demand forecasting module uses the N-BEATS model to capture long-range dependencies in time series data. It takes historical charging data, time-period characteristics (such as peak or off-peak hours), and weather data as input to obtain the charging demand distribution for the next time period t, including the expected number of vehicles connected and the total charging power for each time period. The specific steps for analyzing the long-range dependencies in the preprocessed charging data using the N-BEATS model are as follows:

[0069] The specific steps for analyzing the long-range dependencies of time series in preprocessed charging data using the N-BEATS model are as follows:

[0070] Map the preprocessed historical charging data to input tensors acceptable to the model.

[0071] An N-BEATS deep residual network was built, and multiple stacked fully connected layer modules were configured to capture the trend and seasonal features of the time series.

[0072] A self-supervised learning paradigm is adopted, which uses an encoder-decoder structure to predict the charging demand for multiple future time steps from the input tensor. The cyclic loss function is used to optimize the model's ability to capture long-sequence dependencies and learn the intrinsic structure of the time series.

[0073] By extracting long-term change patterns through a trend generator and capturing short-term cyclical patterns through a seasonal generator, the preliminary distribution of charging demand within the next time t is obtained by superimposing these patterns.

[0074] Multiple independent N-BEATS model instances are trained, and differentiated training is performed based on the preliminary distribution of charging demand in the future time t. The fused instances are obtained by fusion through Bagging.

[0075] Dropout integration is used to perform multiple forward propagations of the fused instance to generate multiple prediction samples. The probability density function of charging demand is generated by statistical analysis of the prediction samples, and the confidence interval of the prediction results is calculated to quantify the prediction uncertainty.

[0076] By using an online learning mechanism, the N-BEATS model parameters are adjusted based on real-time charging data to obtain the charging demand distribution at time t in the future.

[0077] The power limit module is used to predict the regional power grid load based on preprocessed charging data using a graph neural network, analyze the spatial coupling relationship between charging piles, and obtain the upper limit of the available power of the power grid at time t in the future.

[0078] The power grid-charging pile topology graph unit is used to set each charging pile in the area as a graph node. The node characteristics include historical load data and rated power.

[0079] The spatial association unit is used to quantify the load coupling strength between charging piles based on the physical connection relationship of the power grid containing the graph nodes, and the edge weights, to obtain a graph structure that represents spatial association.

[0080] The load transfer unit is used to aggregate features through a graph convolutional network. In each layer of the model, each node collects the load features of its neighboring nodes and uses graph convolution operators to achieve message passing, thereby capturing the load transfer effect between different charging piles.

[0081] Spatial dependency node units are used to generate spatial dependency nodes containing the spatial dependency relationship between different charging piles after multi-layer convolution, implying the overall load pattern of the regional power grid.

[0082] The power upper limit unit is used to concatenate the spatial features output by graph convolution with historical time series. It predicts the load value of each charging pile at time t in the future through fully connected layers or temporal convolutional layers. After accumulation, combined with grid capacity constraints, it calculates the dynamic upper limit of available power in the region, providing supply-side constraints for power allocation.

[0083] The power allocation module is used to initially allocate charging power under the constraints of the charging demand distribution at time t in the future and the upper limit of the available power of the power grid at time t in the future, through a fuzzy control algorithm.

[0084] The variable determination unit is used to determine the input and output variables of fuzzy control, formulate fuzzy rules, and build a rule base. The input and output variables of fuzzy control are determined by transforming the charging demand distribution at future time t into variables such as charging urgency and charging demand intensity, and transforming the upper limit of available grid power at future time t into variables such as grid friendliness and grid load margin. Simultaneously, the power allocation ratio of each charging pile is set as the output variable.

[0085] The input variable transformation extracts relevant information from the charging demand distribution within the next time point t to determine the variable of charging urgency. If, within the next time point t, the probability of high-power charging demand is high and the remaining battery power of vehicles is generally low, then the charging urgency is high. Conversely, if most vehicles have sufficient battery power and the probability of high-power demand is low, then the charging urgency is low. By comprehensively considering the distribution of charging demand in this way and quantifying it as the fuzzy variable of charging urgency, the urgency of users for fast charging can be intuitively reflected.

[0086] Charging demand intensity is also determined based on the distribution of charging demand, focusing on the overall scale of charging demand. It considers factors such as the number of vehicles expected to connect to charging stations within a future time t, and the sum of the expected charging power for each vehicle. If many vehicles connect and the total expected charging power is close to or exceeds the total power supply capacity of charging stations in the area, the charging demand intensity is high. Conversely, if few vehicles connect and the total expected power is low, the charging demand intensity is low. This variable helps to grasp the overall level of electricity demand.

[0087] Grid friendliness is derived from the upper limit of the grid's available power at future time t. A higher upper limit of available power means the grid has a larger power supply margin and a stronger capacity to accept charging power, resulting in higher grid friendliness. Conversely, when the upper limit of available power is low, close to or lower than the current total allocated power, the grid has limited space to accept new charging power, resulting in lower grid friendliness. Grid friendliness reflects the grid's capacity to handle charging power and its level of friendliness.

[0088] The grid load margin is derived from the upper limit of the grid's available power. It is calculated by dividing the difference between the current upper limit of the grid's available power and the full-load operating power of all charging piles in the area by the full-load operating power, resulting in a proportional value that characterizes the grid load margin. For example, a large difference and a high proportional value indicate a large grid load margin, meaning the grid has sufficient capacity to handle new charging loads. A small difference and a low proportional value indicate a small grid load margin, suggesting a heavy grid load, approaching its capacity limit, and requiring careful allocation of charging power to prevent overload.

[0089] The output variable is set based on the power allocation ratio of each charging pile, with a value ranging from 0 to 1. A power allocation ratio of 0.8 for a certain charging pile means that the power currently allocated to it is 80% of its rated power. By adjusting this ratio, the charging power of each charging pile can be dynamically adjusted to adapt to different charging demands and grid conditions.

[0090] Based on experience and actual conditions, fuzzy rules are formulated to build a comprehensive rule base, so that there are corresponding power allocation guidelines for different situations.

[0091] The rules for achieving reasonable dynamic adjustment of charging power are formulated by taking into account multiple factors. These rules are based on past experience and summaries of various situations observed during actual operation.

[0092] In specific embodiments, when the urgency of charging is high and the grid load margin is low, the power allocation ratio of the corresponding charging pile should be appropriately reduced; when the charging demand intensity is moderate and the grid friendliness is high, the power allocation ratio can be increased accordingly.

[0093] From the perspective of battery state of charge (SOC), a low SOC often indicates insufficient battery power and an urgent need for charging. Based on experience, a fuzzy rule can be established: when the SOC is low, the battery temperature is within a suitable range, and the grid load is within acceptable limits, the charging power can be set to a higher level for rapid charging. However, if the battery temperature is too high, even if the SOC is not yet high, for battery safety and lifespan considerations, the corresponding rule might be to appropriately reduce the charging power when the battery temperature is high, regardless of its SOC range, to prevent damage during charging due to overheating.

[0094] Building a comprehensive rule base requires combining various factors that may influence charging power allocation to different degrees and determining corresponding power allocation guidelines. Besides the examples mentioned above, the impact of factors such as different time periods, regions, and seasons on variables like charging urgency, grid load margin, charging demand intensity, and grid friendliness can be considered. Reasonable power allocation rules should be formulated for each combination. For instance, during off-peak hours at night, grid load margin is usually higher, and charging urgency is relatively lower. Even if charging demand intensity is slightly higher, the power allocation ratio of charging piles can be appropriately increased to fully utilize off-peak electricity resources, reduce user charging costs, and not affect grid stability. By comprehensively considering these factors and formulating rules, corresponding power allocation guidelines can be found in the rule base for various practical scenarios, providing a comprehensive basis for the dynamic adjustment of charging power.

[0095] The power allocation unit is used to perform fuzzy reasoning based on input and output variables and a rule base. For the current state of the input variables, it infers the corresponding preliminary conclusion of power allocation according to the rules in the rule base and obtains the fuzzy power allocation result.

[0096] The fuzzy defuzzification unit is used to convert the fuzzy power allocation result into the power allocation value of each charging pile using the centroid method, thereby completing the reasonable allocation of charging power.

[0097] The membership subunit is used to determine the fuzzy set based on the rule base rules and the actual input variables. In a specific embodiment, the fuzzy set regarding power allocation is inferred based on factors such as battery SOC and temperature. The "low power" fuzzy subset corresponds to situations such as high battery SOC, battery temperature within the normal but slightly high range, and high grid load, and has a corresponding membership function to describe the degree of conformity of different power values ​​in this subset. The "medium power" and "high power" fuzzy subsets also have their own membership functions set based on different charging scenario conditions.

[0098] The membership function is expressed as follows:

[0099]

[0100] In the formula, a and b are parameters that determine the range of the fuzzy set, and x is the battery state of charge.

[0101] The subset parameter calculation subunit is used to calculate the membership degree corresponding to different power values ​​in each fuzzy subset according to the membership degree function. In a specific embodiment, when analyzing the power distribution of a charging pile, within the 20-50 kW range corresponding to the "medium power" subset, based on the established membership degree function and the current charging-related conditions, specific values ​​such as a membership degree of 0.8 at 30 kW and a membership degree of 0.6 at 40 kW are calculated.

[0102] The clear power value calculation subunit is used to substitute the power values ​​and membership degrees of each fuzzy subset into the centroid method calculation formula to calculate the clear power allocation value. In a specific embodiment, if the calculated defuzzified power value of a certain charging pile is 35 kilowatts, it means that the charging pile should operate at a power of 35 kilowatts according to the reasonable allocation requirements.

[0103] The real-time feedback module uses a fuzzy control model to recalculate the power allocation scheme, update the prediction results of the N-BEATS model, and synchronously adjust the input parameters of the fuzzy control.

[0104] The re-prediction unit adjusts the computational logic of the N-BEATS model to obtain a new prediction of the charging demand distribution at time t in the future. In a specific embodiment, the specific state of charge (SOC) values ​​of the battery are mapped to fuzzy set concepts such as "low," "medium," and "high," battery temperature is mapped to fuzzy categories such as "low temperature," "suitable temperature," and "high temperature," and grid load is mapped to conditions such as "low load," "medium load," and "high load." Then, reasoning is performed according to the rules in the rule base to deduce a new preliminary conclusion on power allocation. This conclusion is initially fuzzy, and then defuzzification methods such as the centroid method are used to transform it into a clear and accurate power allocation scheme to adapt to various dynamic changes in actual charging scenarios.

[0105] The re-prediction unit adjusts the computational logic of the N-BEATS model to obtain a new prediction of the charging demand distribution at time t in the future. The N-BEATS model was originally used to predict charging power and related factors. It uses historical data and a predetermined algorithm to infer future charging power trends and battery state changes. However, as actual conditions change, previous predictions may no longer be accurate. Therefore, the power allocation scheme recalculated through a fuzzy control model is used as a new reference and fed back into the N-BEATS model. This adjusts its internal parameters and computational logic, enabling the N-BEATS model to re-predict subsequent battery charging states and power demands based on the new power allocation, making its predictions more aligned with actual development trends and more accurate and valuable.

[0106] The feedback adjustment unit is used to input the newly predicted charging demand distribution within the next time t into the fuzzy control algorithm again, so as to obtain the power allocation scheme in the next round of calculation. In a specific embodiment, after the battery has been charging for a period of time, the state of charge changes, or the grid load changes due to the connection or disconnection of other electrical devices. Therefore, the newly obtained accurate parameters such as battery state of charge, battery temperature, and grid load need to be input into the fuzzy control model again to replace the old parameters. In this way, the fuzzy control model can accurately infer the appropriate power allocation scheme again in the next round of calculation based on the latest actual situation, thus forming a good mechanism of continuous looping, real-time feedback and dynamic adjustment. This ensures that the entire charging power allocation and related prediction work can continue to be carried out continuously and accurately, meeting the needs for reasonable power allocation and accurate prediction during the charging process.

[0107] The real-time monitoring module displays the real-time power distribution curves of each charging pile, vehicle charging progress, and dynamic graphs of grid load. It also uses color coding to distinguish charging priorities, with red indicating emergency charging and green indicating normal charging, and generates predictive analysis reports.

[0108] The system displays real-time power distribution curves for each charging station. Throughout the entire charging system operation, the real-time monitoring module continuously collects power distribution data from each charging station. For each charging station, it plots a real-time changing curve with time on the horizontal axis and power distribution value on the vertical axis. These curves allow maintenance personnel or relevant managers to visually observe the power distribution of each charging station at different times, such as whether there are sudden increases or decreases in power, or whether the power distribution is stable. This helps to promptly detect potential anomalies at the charging stations, such as uneven power distribution or excessive power fluctuations affecting charging efficiency. Furthermore, the curve trends can be analyzed to understand the usage patterns of each charging station over different time periods, providing a basis for subsequent resource allocation and equipment maintenance.

[0109] The module displays vehicle charging progress by acquiring real-time charging status information for each vehicle, including current battery level, elapsed charging time, and estimated time remaining for a full charge. This information is displayed as a percentage of battery charge, such as 50% charge, or as a progress bar. By showing the charging progress, car owners can clearly understand their vehicle's charging status and plan their subsequent trips accordingly. Furthermore, from a system management perspective, it allows for proactive planning to handle peak and off-peak charging periods based on the overall charging progress, ensuring efficient and orderly charging services.

[0110] This module displays a dynamic graph of the power grid load. It monitors the power grid load in real time, integrates and processes relevant data, and presents it in the form of a dynamic graph. The graph can be a line chart showing the load trend over time, or a bar chart comparing load levels at different times. By observing this dynamic graph, users can clearly understand the load pressure on the power grid at different points in time, predict potential overload risks, and take timely measures, such as adjusting the power distribution of charging stations, to prevent malfunctions caused by excessive grid load and ensure the safe and stable operation of the entire charging system and the power grid.

[0111] Color coding distinguishes charging priorities. To more clearly differentiate the charging priorities of different vehicles, the real-time monitoring module uses color coding. Red is specifically used to indicate vehicles requiring emergency charging, often due to urgent travel needs or low battery levels. Green indicates vehicles undergoing normal charging, meaning they do not require emergency charging and can be charged according to the normal charging sequence and pace. This intuitive color coding allows for quick identification of which vehicles need priority charging, whether viewed on the monitoring screen or during on-site inspections by staff. This helps in the rational allocation of charging resources and improves overall service satisfaction.

[0112] The real-time monitoring module also features predictive analysis reports. Based on collected historical and real-time data, it uses data analysis and prediction algorithms to generate reports covering various aspects. For example, it can predict charging demand over a future period, analyzing potential charging peaks and troughs. It can also estimate battery wear and tear, helping car owners understand the impact of charging behavior on battery life. Furthermore, it can plot grid load trends, revealing potential changes in grid load pressure. The reports may also provide comparative analyses of different power allocation strategies and charging modes, such as the advantages and disadvantages of "fast charging mode" versus "economic charging mode," providing strong data support and reference for decision-makers in formulating charging management strategies and optimizing resource allocation.

[0113] In summary, this embodiment provides an AI-based dynamic adjustment and intelligent allocation system for charging power. It uses an N-BEATS model to predict the charging demand distribution at time t and a graph neural network to predict the regional power grid load, thus obtaining the upper limit of the available power of the power grid at time t. A fuzzy control algorithm is then used to rationally allocate the charging power. The N-BEATS model can deeply analyze the long-distance dependencies hidden in the pre-processed charging data, thereby accurately predicting the charging demand distribution at time t, providing an important time-dimensional reference for subsequent power allocation. The graph neural network excels at capturing spatial structure information in the data. By analyzing the spatial coupling relationships between charging piles, it accurately predicts the regional power grid load based on the pre-processed charging data, thereby deriving the upper limit of the available power of the power grid. This provides crucial spatial dimension support for considering the collaborative operation of multiple charging piles and the overall carrying capacity of the power grid. The fuzzy control algorithm, with its effective processing capability for fuzzy and uncertain information, rationally allocates power under the constraints of charging demand distribution and the upper limit of available power of the power grid, both of which have certain fuzzy characteristics. Whether in the initial allocation or subsequent real-time feedback adjustments, it can effectively balance various complex factors, ensuring the scientific and rational nature of the power allocation.

[0114] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI-based dynamic adjustment and intelligent allocation system for charging power, characterized in that, include: The data acquisition module is used to collect charging data, including real-time charging data and historical charging data; The data preprocessing module is used to preprocess the charging data to obtain preprocessed charging data; The electricity consumption prediction module is used to analyze the long-distance dependence of the time series in the preprocessed charging data using the N-BEATS model to obtain the charging demand distribution in the future time t. The power limit module includes: a charging pile topology graph unit, used to set each charging pile in the area as a graph node; a spatial association unit, used to connect edges according to the physical connection relationship of the power grid and quantify the load coupling strength to construct a graph structure representing spatial association; a load conduction unit, used to perform feature aggregation on the graph structure representing spatial association, use graph convolution operators for message passing, and obtain the load conduction effect between different charging piles; a spatially dependent node unit, used to generate spatially dependent nodes containing the load conduction effect between different charging piles after convolution; and a power limit unit, used to concatenate the spatial features of the spatially dependent nodes with historical time series, predict the load value of each charging pile at time t in the future through a time-series convolutional layer, and calculate the dynamic limit of available power in the area based on the load value. The power allocation module is used to initially allocate charging power using a fuzzy control algorithm, under the constraints of the charging demand distribution in the future time t and the upper limit of the available power of the power grid in the future time t. The real-time feedback module uses a fuzzy control algorithm to recalculate the power allocation scheme, update the prediction results of the N-BEATS model, and obtain a clear charging power allocation.

2. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 1, characterized in that, The data acquisition module includes: The vehicle battery data acquisition unit is used to collect information such as current battery level, battery health status, remaining charging capacity, and charging interface type. The charging pile data acquisition unit is used to collect rated power, current load, and available power threshold. The power grid data acquisition unit is used to collect real-time voltage, real-time current, power factor, regional load rate, and electricity price during specific time periods. The user data acquisition unit is used to collect the target charging capacity, expected completion time, and charging priority. The environmental data acquisition unit is used to collect air temperature and humidity.

3. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 1, characterized in that, The data preprocessing module includes: The feature extraction unit is used to extract key features from the charging data; A data standardization unit is used to standardize the battery's maximum acceptable charging power and available power resources based on the historical charging data. The data cleaning unit is used to check whether there are any abnormalities in the charging data and to correct them.

4. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 1, characterized in that, The processing steps of the electricity consumption forecasting module include: Map the historical charging data in the preprocessed charging data to an input tensor; Construct an N-BEATS deep residual network model; The ability of the N-BEATS deep residual network model to capture long-sequence dependencies is optimized by using the cyclic loss function, and the intrinsic structure of time series is learned. The long-term variation pattern and short-term periodic pattern are extracted by the trend generator and seasonal generator of the N-BEATS deep residual network model, respectively. The long-term variation pattern and short-term periodic pattern are superimposed to obtain the preliminary distribution of charging demand in the future time t. Multiple independent N-BEATS model instances are trained, and differentiated training is performed based on the preliminary distribution of charging demand in the future time t. The preliminary distribution of charging demand in the future time t after differentiated training is fused to obtain a fused instance. Dropout integration is used to perform multiple forward propagations on the fusion instance to generate multiple prediction samples. The probability density function of the charging demand generated by the prediction samples is statistically analyzed, the confidence interval of the prediction results is calculated, and the prediction uncertainty is quantified. By using an online learning mechanism, the N-BEATS model parameters are adjusted based on the real-time charging data to obtain the charging demand distribution within the future time t.

5. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 1, characterized in that, The power distribution module includes: The variable determination unit is used to determine the input and output variables of fuzzy control, formulate fuzzy rules, and build a rule base; The power allocation unit is used to perform fuzzy reasoning based on the input and output variables and the rule base, infer a preliminary conclusion on power allocation according to the rules in the rule base, and obtain a fuzzy power allocation result. The fuzzy de-fuzzification unit is used to convert the fuzzified power allocation result into power allocation values ​​for each charging pile using the centroid method.

6. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 5, characterized in that, The blur-reducing unit includes: Membership subunits are used to determine fuzzy sets based on the rules of the rule base and the actual input variables; The subset parameter calculation subunit is used to calculate the membership degree corresponding to different power values ​​in each fuzzy subset using the membership degree function; The clear power value calculation subunit is used to substitute the power values ​​and membership degrees of each fuzzy subset into the centroid method calculation formula to calculate the clear power allocation value.

7. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 6, characterized in that, The membership function is expressed as follows: In the formula, a and b are parameters that determine the range of the fuzzy set, and x is the battery state of charge.

8. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 6, characterized in that, The real-time feedback module includes: The result update unit is used to update the prediction results of the N-BEATS model using the initially allocated charging power. The re-prediction unit is used to adjust the computational logic of the N-BEATS model to obtain a new prediction of the charging demand distribution in the future time t. The feedback adjustment unit is used to input the newly predicted charging demand distribution within the next time t into the fuzzy control algorithm again, so as to obtain the power allocation scheme in the next round of calculation.

9. The AI-based dynamic adjustment and intelligent allocation system for charging power according to claim 1, characterized in that, It also includes a real-time monitoring module, which is used to: collect power allocation data related to charging piles and display the real-time power allocation curve of each charging pile; monitor the grid load in real time, display the grid load dynamic graph, and use color coding to distinguish charging priorities.

Images