Charging pile load prediction method and system

Abstract

The invention discloses a charging pile load prediction method and system, and relates to the technical field of power grid cooperative control. Load prediction is no longer limited to pile end historical data by capturing bidirectional data interaction between an electric vehicle and a charging pile in real time and combining a dynamic battery fingerprint database based on a prediction model constructed based on a vehicle pile interaction dynamic time sequence; the dynamic prediction engine accurately adapts to the power calculation of the constant current / constant voltage stage according to the SoC, SOH and temperature parameters of the battery, and the generated load curve can reflect the influence of the battery state change on the charging behavior in real time, so that the prediction result better fits the actual charging scene, the accurate load prediction is provided for the power grid dispatching, and the power grid dispatching efficiency is improved. Therefore, the power grid can adjust power distribution in advance according to prediction, meanwhile, accurate load reference is provided for charging pile operation, and deep fusion and collaborative response of vehicle, pile and network data are realized.

Description

Technical Field

[0001] This invention relates to the field of power grid collaborative control technology, specifically to a method and system for predicting the load of charging piles. Background Technology

[0002] With the energy transition, the penetration rate of electric vehicles continues to rise, and the construction scale of public and private charging piles is growing explosively. The proportion of charging load in the total grid load is increasing year by year. Especially during peak commuting hours and holidays, the concentration of charging behavior often leads to increased fluctuations in regional grid load, posing new challenges to the safe and stable operation of the grid. At the same time, users' requirements for charging efficiency and charging experience are constantly increasing. How to achieve accurate prediction of charging load and coordinated grid dispatch while meeting user needs has become a key issue for the development of the new energy vehicle industry.

[0003] Existing technologies have significant limitations in charging load prediction and collaborative control. In terms of prediction, traditional methods rely heavily on historical load data of charging piles or single environmental factors, failing to effectively integrate the battery status of electric vehicles with the dynamic interaction between vehicles and charging piles. This results in insufficient response to key variables affecting charging behavior, such as battery aging and fast charging modes. In terms of data utilization, there are significant data barriers among charging pile operators, and a lack of cross-entity data collaboration mechanisms under privacy protection makes it difficult to form accurate regional prediction models. Furthermore, existing control strategies are mostly unidirectional command execution, failing to consider the charging urgency of different vehicles, which can easily lead to conflicts between user experience and power grid safety.

[0004] In summary, current charging load forecasting suffers from insufficient accuracy and weak coordination, failing to meet the grid dispatching needs under the large-scale development of new energy vehicles. Therefore, developing a forecasting and control method that can deeply integrate vehicle-charging dynamic data, possess data coordination capabilities with privacy protection, and achieve dynamic balance between grid and vehicle demand has become a pressing technical challenge for the industry. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies by providing a charging pile load prediction method and system. This system establishes a real-time data exchange link between the vehicle and the charging pile via a communication interface, acquires vehicle BMS data in real time, and combines it with a benchmark charging curve template from a cloud-based fingerprint database. A dynamic prediction engine accurately calculates the fast charging coefficient, temperature influence coefficient, and battery health influence coefficient, constructing a realistic charging load prediction model that fully reflects the impact of battery state changes on charging behavior. The generated energy demand curve accurately presents future charging load trends, providing a reliable basis for grid dispatching and enabling the grid to allocate power resources in advance, ensuring the stable operation of the power system. It also provides data support for charging pile operation and management, facilitating the rational planning of equipment maintenance and operation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a charging pile load prediction method, the specific steps of which are as follows:

[0007] S100: The charging pile obtains vehicle battery management system (BMS) data in real time via the vehicle-to-pile communication protocol. Simultaneously, the cloud-based fingerprint database returns a reference charging curve template matching the battery model. This reference charging curve template includes a power function P for constant current / constant voltage stages. base (t,SoC);

[0008] S200 and the dynamic prediction engine receive the baseline charging curve template and BMS data from the cloud fingerprint database in S100. Based on the real-time BMS data, they load dynamic coefficients to adjust the template parameters and calculate the predicted power P. predict (t);

[0009] S300, monitor actual charging power P real (t), calculation error ΔP=|P predict -P real When ΔP > 10% and lasts for 5 minutes, each charging station will encrypt and upload its error feature vector. After removing outliers in the cloud, the vectors will be grouped and aggregated by battery model to form the error feature set V for that battery model. er ;

[0010] S400 uses aggregated data to train a neural network to update the curve correction model in the cloud and sends the updated template differential to the charging pile of the same model. After receiving the template, the charging pile immediately switches the prediction engine parameters and stores the old template in the historical version library.

[0011] The S500 predicts the total load curve of the region output by the engine and converts it into grid communication protocol instructions. It then adjusts the power based on the actual grid load and the SoC value of the charging pile group.

[0012] Furthermore, in S100, the charging pile establishes a real-time data interaction link with the vehicle battery management system (BMS) through a communication interface, sends a status read command to the BMS, and requests to obtain real-time data items including: the current remaining battery power SoC value, battery health SOH, battery model, and simultaneously collects battery temperature data T and the vehicle's unique identification code.

[0013] The charging pile will standardize the collected battery model, remove redundant characters to generate a unique model code, associate it with the vehicle's unique identification code and the current charging session, bind the battery data packet to the charging pile port number and timestamp t, and send it to the pre-stored cloud fingerprint database through an encrypted transmission channel.

[0014] After receiving the model code, the cloud-based fingerprint database matches it with the corresponding reference charging curve template, which includes the constant current stage power function P. base_cc (t,SoC) and the power function P during the constant voltage stage base_cv (t,SoC);

[0015] The power function P during the constant current stage base_cc (t,SoC) satisfies P base_cc (t,SoC)=P rated ×f(SoC), where P rated The rated power of the charging pile is f(SoC), which is a linear function of SoC. When SoC is between 0% and 80%, f(SoC) = 1.0, and when SoC is between 80% and 95%, f(SoC) = 0.5.

[0016] The power function P during the constant voltage stage base_cv (t,SoC) satisfies P base_cv (t,SoC)=k×U rated ×I(t), where k is the conversion efficiency coefficient preset according to the battery model, with a value of 0.92-0.95, U rated Let I(t) be the rated voltage of the battery, and I(t) be the decreasing current function during the constant voltage phase. I max t0 is the maximum charging current at the start of the constant voltage phase, and t0 is the cumulative charging time from the start of the constant voltage phase.

[0017] The cloud will encapsulate a benchmark charging curve template containing the power function of the constant current / constant voltage stage and send it to the corresponding charging pile. The template data includes a time axis, the power value at each time point, and a description of the function parameters.

[0018] Furthermore, the dynamic prediction engine in S200 receives the benchmark charging curve template from the cloud fingerprint database in S100, and parses the constant current stage power function P in the template. base_cc (t,SoC) and the power function P during the constant voltage stage base_cv (t,SoC), simultaneously acquires BMS data transmitted in real time by the charging pile through the communication link, and follows P predict (t)=P base (t,SoC)×k temp (T)×k SOH (SOH) Calculate and predict the power P within 15 minutes predict (t), where k temp (T) is the temperature influence coefficient, k SOH (SOH) is the battery health impact coefficient, P base (t,SoC) is the power function for the constant current / constant voltage stage.

[0019] Furthermore, in S200, the dynamic prediction engine loads dynamic coefficient adjustment template parameters based on real-time BMS data, including:

[0020] The acquired SoC value is evaluated. When the SoC is less than 30%, the constant current charging stage is initiated, and the preset fast charging coefficient k is enabled. fast That is, P predict (t)=P base (t,SoC)×k temp (T)×k SOH (SOH)×k fast For batteries with SOH ≥ 90%, k fast The value is 1.2-1.3. For batteries with 70% ≤ SOH < 90%, k fast The value is 1.1-1.2. For batteries with SOH < 70%, k fast Values ​​range from 1.0 to 1.1; when SoC ≥ 30%, it enters the constant current stage, k fast The value is 1.0;

[0021] For the temperature influence coefficient k temp (T), when the battery temperature is in the optimal operating range of 25℃-35℃, k temp (T) takes a value of 1.0. When the temperature is below 25℃, k increases by 1℃ for every 1℃ decrease. temp (T) decreases linearly by 0.01. When the temperature is above 35℃, for every 1℃ increase, k temp (T) decreases linearly by 0.02, and k temp (T) The minimum value is not less than 0.8;

[0022] The influence coefficient k of battery health SOH (SOH), when SOH ≥ 80%, k SOH (SOH) = 0.8 + 0.002 × (SOH - 80), where k is less than 80%. SOH (SOH) = 0.8;

[0023] Based on the calculated predicted power, a power demand curve for 15 minutes is generated according to the timestamp t, and associated with the corresponding charging pile port number and vehicle unique identification code, and synchronized to the cloud.

[0024] Furthermore, in S300, the charging pile collects its own actual output charging power P in real time. real (t), and associate it with the corresponding timestamp, vehicle unique identifier and charging pile port number;

[0025] Call the concurrent predicted power data P generated in S200 predict (t), according to ΔP=|P predict-P real | Calculate the power error value every minute, and record the corresponding SoC value, battery temperature and SOH health status for each calculation;

[0026] The error value is monitored for 5 consecutive minutes. When ΔP > 10% is met for 5 consecutive minutes, the error data upload mechanism is triggered.

[0027] The charging pile packages the error data that triggers the upload, forming an error feature vector V. e = [ΔP,SoC,T,SOH], where each parameter is accompanied by a corresponding timestamp and is uploaded to the cloud via an encrypted transmission channel;

[0028] The cloud receives encrypted error feature vectors uploaded by each charging pile, cleans the data, removes outliers, and groups the cleaned valid error feature vectors according to the battery model obtained from S100. Error feature vectors of the same battery model are then aggregated to generate an error feature set V for that battery model. er .

[0029] Furthermore, in S500, the cloud calls a preset neural network architecture to train a curve correction model. This architecture uses three fully connected layers, including an input layer, a hidden layer, and an output layer, to generate an error feature set V for the same battery model. er As a training dataset, it is divided into a training set and a validation set in an 8:2 ratio;

[0030] The curve correction model is trained using the training set, and the validation set is input to complete the training. The new fast-charging coefficient k is then output. new-fast ;

[0031] Based on k new-fast The power function P in the constant current stage of the reference charging curve template base_cc (t,SoC) is corrected using the formula P. new_cc (t,SoC)=P base_cc (t,SoC)×(k new-fast / k fast ), where k fast For the original fast charging coefficient in the template, a differential update package is generated. The differential update package only contains the difference data from the old template.

[0032] The cloud matches the corresponding list of charging piles based on the battery model and sends the differential update package to the charging piles corresponding to the same battery model through an encrypted transmission channel.

[0033] After receiving the differential update package, the charging pile automatically merges it with the old template stored locally to generate a new template, replacing the current template parameters in the dynamic prediction engine. At the same time, the old template is stored in the historical version library according to the naming rules of battery model and update timestamp, and a successful update receipt is sent to the cloud.

[0034] Furthermore, in S500, the prediction engine aggregates the energy demand curves of all charging piles within the area over a 15-minute period, aligns them according to the time axis, and then uses L... total (t)=∑P predict (t) Calculate the total load curve for the region;

[0035] The cloud platform converts the regional total load curve into grid communication protocol instructions, the instruction format of which includes start time, duration, required power, and maximum power ramp rate.

[0036] The power grid dispatching system receives protocol instructions and monitors the actual total load of charging piles in the area in real time. When the actual load exceeds a preset safety threshold for 30 consecutive seconds, it sends a power reduction command to the charging piles. The safety threshold is 90% of the maximum carrying capacity of the regional power grid. The power reduction command is as follows:

[0037] After receiving the power reduction command, the charging pile queries the SoC value of the currently charging vehicle through the established vehicle-to-pile communication link. When the vehicle is charging and the SoC is ≥70%, the output power is reduced to 80% of the current actual power. When the vehicle's SoC is ≤30%, the output power is reduced to 90% of the current actual power.

[0038] After the charging pile completes the power adjustment, it reports the actual power value and adjustment completion time to the power grid dispatch system. The power grid dispatch system monitors the total load of the area after the adjustment in real time. When the total load falls below the safety threshold, the power reduction state is lifted, allowing the charging pile to resume normal power output.

[0039] On the other hand, the charging pile load prediction system consists of: a vehicle-charging pile data interaction module, a cloud fingerprint database module, a dynamic prediction engine module, a federated learning update module, and a power grid collaborative control module.

[0040] The vehicle-charging pile data interaction module is used to establish a real-time communication link between the charging pile and the vehicle BMS, obtain real-time BMS data such as SoC value, battery temperature, SOH health, and battery model, and send the standardized battery model data to the cloud fingerprint library module. At the same time, it receives the benchmark charging curve template sent from the cloud.

[0041] The cloud fingerprint database module pre-stores the benchmark charging curve templates corresponding to different battery models. After receiving the battery model code sent by the vehicle-charging pile data interaction module, it matches and sends out the corresponding template, and at the same time stores the historical template versions of each battery model.

[0042] The dynamic prediction engine module receives templates from the cloud fingerprint library module and real-time data transmitted from the vehicle-charging data interaction module, parses the template function and determines the SoC value, activates the corresponding fast charging coefficient, and generates the predicted power and the energy demand curve within 15 minutes.

[0043] The federated learning update module monitors the error between the actual charging power and the predicted power, encrypts and uploads the error feature vector to the cloud, receives the curve correction model and updated template generated by the cloud through neural network training, replaces the local template and stores the old template in the historical version library.

[0044] The power grid collaborative control module summarizes the regional total load curve generated by the dynamic prediction engine module, converts it into power grid communication protocol instructions, and controls the charging pile to adjust the output power according to the vehicle SoC value.

[0045] Compared with existing technologies, this charging pile load prediction method and system have the following advantages:

[0046] I. This invention is based on a predictive model constructed from the dynamic time sequence of vehicle-charging pile interaction. By capturing the bidirectional data interaction between electric vehicles and charging piles in real time and combining it with a dynamic battery fingerprint database, load prediction is no longer limited to historical data at the charging pile. The dynamic prediction engine accurately adapts to the power calculation of the constant current / constant voltage stage based on the parameters of battery SoC, SOH, and temperature. The generated load curve can reflect the impact of battery state changes on charging behavior in real time, making the prediction results more consistent with the actual charging scenario. This provides accurate load prediction for grid dispatch, enabling the grid to adjust power distribution in advance based on the prediction. At the same time, it provides accurate load reference for charging pile operation, realizing deep integration and coordinated response of vehicle, charging pile, and grid data.

[0047] Second, this invention utilizes a federated learning-driven fingerprint database evolution mechanism to achieve effective aggregation and template optimization of cross-charging pile data by encrypting and transmitting error feature vectors and collaboratively training curve correction models, while protecting the data privacy of each charging pile. It also dynamically adjusts the charging power based on the total regional load and vehicle SoC status, ensuring grid security while taking into account the charging needs of low SoC vehicles. This allows the model to continuously adapt to changes in battery characteristics. Furthermore, it constructs a flexible interaction mechanism between the grid and charging piles, realizing an intelligent upgrade of load forecasting and control, and promoting the development of charging systems towards a more efficient and collaborative direction.

[0048] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0050] Figure 1 A flowchart illustrating the operation of the charging pile load prediction method;

[0051] Figure 2 This is a step-by-step diagram of the charging pile load prediction method. Detailed Implementation

[0052] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0053] Example 1

[0054] This embodiment uses a regional charging pile cluster as an application scenario to elaborate on the overall implementation steps and working principle of the charging pile load prediction method, such as... Figure 2 As shown, this method obtains battery data through real-time interaction between vehicles and charging piles, dynamically generates predicted power by combining cloud-based benchmark curve templates, continuously optimizes the model based on error feedback, and ultimately achieves regional load collaborative control. It fully demonstrates the technical characteristics of vehicle, charging pile, and grid data fusion and dynamic correction, and provides a solution for the safe and stable operation of the power grid and efficient scheduling of charging piles.

[0055] First, the data acquisition and benchmark template matching stage (S100) is entered. A vehicle-to-charging pile data interaction link is established, and a matching benchmark charging curve template is obtained, providing basic data and an initial model for subsequent predictions. After a vehicle connects, the charging pile immediately establishes a real-time data interaction link with the vehicle's Battery Management System (BMS) through a preset vehicle-to-charging pile communication protocol. The communication interface sends standardized status read commands to the BMS, specifying the requested real-time data items, including the battery's current remaining charge (SoC), battery health status (SOH), and battery model. Simultaneously, battery temperature data (T) and the vehicle's unique identification code are collected. After obtaining the battery model, the charging pile standardizes it, removing redundant characters and generating a unique model code to ensure that the same model battery reported by different vehicles can be uniformly identified. The charging pile associates the vehicle's unique identification code with the current charging session, binds the battery data packet to its own port number and timestamp (t), and sends it to the cloud fingerprint database through an encrypted transmission channel. Upon receiving the model code, the cloud fingerprint database performs precise matching in a pre-stored template library to find the corresponding benchmark charging curve template, which includes the constant current stage power function P. base_cc (t,SoC) and the power function P during the constant voltage stage base_cv (t,SoC), where the constant current stage power function P base_cc (t,SoC)=P rated ×f(SoC), P rated The rated power of the charging pile is given by f(SoC), which is a linear function of SoC. When SoC is between 0% and 80%, f(SoC) = 1.0; when SoC is between 80% and 95%, f(SoC) = 0.5. The constant voltage stage power function P... base_cv (t,SoC)=k×U rated ×I(t), k is the preset conversion efficiency coefficient based on the battery model (value ranges from 0.92 to 0.95), U rated This is the battery's rated voltage. (I max t0 is the maximum charging current at the start of the constant voltage phase and t0 is the cumulative charging time during the constant voltage phase. The cloud will encapsulate the benchmark charging curve template containing these two functions and send it to the corresponding charging pile. The template also includes a time axis, power values ​​at each time point, and function parameter descriptions to facilitate the analysis and use of the charging pile.

[0056] Then, the dynamic power prediction stage (S200) begins. After receiving the baseline charging curve template from the cloud, the dynamic prediction engine first parses the constant current stage power function P in the template. base_cc (t,SoC) and the power function P during the constant voltage stage base_cv(t,SoC) defines the power calculation logic for the two stages. Simultaneously, the engine continuously receives BMS data transmitted in real-time from the charging pile via a communication link. This data includes real-time updates of SoC, SOH, T, etc., providing a basis for dynamically adjusting prediction parameters. The predicted power is calculated based on formula P. predict (t)=P base (t,SoC)×k temp (T)×k SOH (SOH), where P base (t,SoC) is the reference power function (the function for either constant current or constant voltage stages is selected based on the SoC value), k temp (T) is the temperature influence coefficient, k SOH (SOH) is the battery health impact coefficient. The dynamic prediction engine determines whether to enable the fast charging coefficient k based on the SoC value. fast When the SoC is below 30%, the battery is in a fast-charging state, and the engine activates the preset k. fast At this point, the prediction formula becomes P predict (t)=P base (t,SoC)×k temp (T)×k SOH (SOH)×k fast When the SoC is greater than 30%, to avoid damage to the battery from excessive fast charging, k fast A value of 1.0 means fast charging gain is not enabled, affecting the temperature influence coefficient k. temp (T) is set based on the battery's optimal operating temperature range (25℃-35℃), within which the battery performance is stable, k temp (T) is set to 1.0. When the temperature is below 25℃, the internal resistance of the battery increases and the charging efficiency decreases. For every 1℃ decrease in temperature, k temp (T) decreases linearly by 0.01. When the temperature exceeds 35℃, the battery faces the risk of thermal runaway, requiring a significant reduction in charging power. For every 1℃ increase, k temp (T) decreases linearly by 0.02, and the minimum value is not less than 0.8 to ensure basic charging needs. The battery health impact coefficient k SOH (SOH) is adjusted according to the SOH value. When SOH ≥ 80%, k SOH (SOH) = 0.8 + 0.002 × (SOH - 80), where k is less than 80%. SOH (SOH) = 0.8, meaning the higher the SOH, the larger the coefficient, reflecting better charging efficiency of a healthy battery. When SOH < 80%, battery performance declines significantly. SOH With (SOH) fixed at 0.8, the dynamic prediction engine calculates and predicts the power P for the next 15 minutes based on the adjusted parameters. predict(t) generates the corresponding power demand curve according to the timestamp t, and synchronizes the curve with the charging pile port number and the vehicle's unique identification code to the cloud to provide data support for the calculation of the regional total load.

[0057] Subsequently, the process enters the error monitoring and feature aggregation stage (S300). During the charging process, the charging pile collects its own actual output charging power P in real time. real (t) is then associated with and stored along with the corresponding timestamp, vehicle unique identifier, and charging pile port number to ensure data traceability. Simultaneously, the concurrent predicted power data P generated in S200 is retrieved. predict (t), according to the error formula ΔP=|P predict -P real The power error value is calculated minute by minute, and the corresponding SoC value, battery temperature T, and SOH health level are recorded for each calculation. Error monitoring adopts a continuous 5-minute judgment mechanism. When ΔP > 10% for 5 consecutive minutes, it indicates that there is a large deviation between the current prediction model and the actual situation, triggering the error data upload mechanism. The charging pile packages the triggered error data to form an error feature vector V. e =[ΔP,SoC,T,SOH], each parameter is accompanied by a corresponding timestamp, and is uploaded to the cloud through an encrypted transmission channel. After receiving the encrypted error feature vectors uploaded by each charging pile, the cloud performs data cleaning and removes outliers. The cleaned effective error feature vectors are grouped according to the battery model obtained in S100, and the error feature vectors of the same battery model are aggregated to generate the error feature set V of that battery model. er This provides a high-quality dataset for subsequent model training.

[0058] Secondly, the template update and parameter synchronization phase (S400) begins, where the cloud utilizes the aggregated set of error features V for the same battery model. er The training process involves a curve correction model constructed from a neural network. This neural network employs a three-layer fully connected architecture (input layer, hidden layer, and output layer). The input layer receives parameters from the error feature vector, the hidden layer performs feature extraction and nonlinear transformation through activation functions, and the output layer generates parameters used to correct the baseline template. During training, the dataset is divided into a training set and a validation set in an 8:2 ratio. The training set is used for iterative optimization of model parameters, while the validation set is used to evaluate model performance and prevent overfitting. After training is complete, a new fast-charging coefficient k is output. new-fast Based on this coefficient, the power function of the constant current stage in the reference charging curve template is corrected, and the correction formula is P. new_cc (t,SoC)=P base_cc (t,SoC)×(k new-fast / k fast ), where k fastThe correction mechanism for the original fast charging coefficient in the template allows the benchmark template to be continuously optimized based on actual errors, improving the prediction accuracy for the same battery model. The corrected template generates a differential update package. The cloud matches the corresponding charging pile list according to the battery model and sends the differential update package to the charging piles of the same model through an encrypted transmission channel. After receiving the differential update package, the charging pile automatically merges it with the old template stored locally to generate a new template, replacing the current template parameters in the dynamic prediction engine to ensure that subsequent predictions use the latest model. At the same time, the old template is stored in the historical version library according to the naming rules of battery model and update timestamp for easy traceability and version rollback. After the charging pile completes the update, it sends an update success receipt to the cloud, forming a closed-loop update confirmation mechanism.

[0059] Finally, the regional load forecasting and power adjustment phase (S500) begins. The forecasting engine aggregates the energy demand curves of all charging piles in the region for the next 15 minutes, aligns them according to the time axis, and then uses formula L... total (t)=∑P predict (t) Calculate the regional total load curve, which reflects the total power demand of the charging pile cluster in the region within the next 15 minutes. The cloud converts the regional total load curve into a grid communication protocol command. The command format includes information such as start time, duration, required power, and maximum power ramp rate to ensure that the grid dispatch system can accurately parse and execute it. After receiving the protocol command, the grid dispatch system monitors the actual total load value of the charging piles in the region in real time. When the actual load exceeds the preset safety threshold (90% of the maximum carrying capacity of the regional grid) for 30 consecutive seconds, it sends a power reduction command to the charging pile. After receiving the power reduction command, the charging pile queries the SoC value of the currently charging vehicle through the vehicle-to-pile communication link and, based on the S... The SoC value implements a differentiated power adjustment strategy: when the vehicle is charging and SoC ≥ 70%, the battery is close to full charge and the demand for charging speed is low, so the output power is reduced to 80% of the current actual power; when the vehicle SoC ≤ 30%, the battery urgently needs charging, so to ensure the user's basic needs, the output power is reduced to 90% of the current actual power. After the charging pile completes the power adjustment, it reports the actual power value and adjustment completion time to the power grid dispatch system. The power grid dispatch system monitors the total load of the area after the adjustment in real time. When the total load falls below the safety threshold, the power reduction state is lifted, allowing the charging pile to resume normal power output, so that the charging load of the entire area returns to a reasonable range.

[0060] In summary, this embodiment achieves accurate prediction and dynamic control of charging pile load by fully executing five stages: data acquisition and benchmark template matching, dynamic power prediction, error monitoring and feature aggregation, template updating and parameter synchronization, and regional load prediction and power adjustment. This process fully utilizes real-time vehicle-charging pile data and cloud-based collaborative optimization mechanisms. Through formulaic parameter calculations and model corrections, the prediction results continuously align with actual charging scenarios, while also considering grid security and user needs. This verifies the feasibility of the charging pile load prediction method and provides strong support for the efficient operation of large-scale charging pile clusters and grid collaborative scheduling.

[0061] Example 2

[0062] like Figure 1 As shown in the figure, this embodiment provides a specific process for predicting the load of charging piles. The steps of the specific process are as follows:

[0063] (1) Real-time acquisition of vehicle battery data

[0064] The charging pile obtains real-time data from the vehicle's BMS (including battery power SoC, SOH, model, and temperature) through the vehicle-to-charging-pile communication protocol.

[0065] After standardizing the battery model, send it to the cloud fingerprint database to request a matching baseline charging curve template for that model.

[0066] (2) Issue the reference charging curve template

[0067] The cloud fingerprint database returns a matching template, which includes power calculation rules for the constant current / constant voltage stage (no formula required, only explanation of the function).

[0068] (3) Dynamically predicting charging power

[0069] The dynamic prediction engine combines real-time BMS data (SoC, temperature, SOH) and template parameters to load dynamic coefficients (such as fast charging coefficient and temperature influence coefficient).

[0070] Adjust the prediction parameters based on the battery status (e.g., enable fast charging if SoC < 30%) to generate a power prediction curve for the next 15 minutes.

[0071] (4) Monitor actual power and error

[0072] The actual output power of the charging pile is collected in real time and compared with the predicted value minute by minute.

[0073] If the error persists for more than 10% for 5 minutes, the error data (including SoC, temperature, and SOH) will be encrypted and uploaded to the cloud.

[0074] (5) Aggregate data and train models in the cloud.

[0075] Cloud-based cleanup of abnormal data, grouping and aggregating error characteristics by battery model.

[0076] The model is trained and corrected using a neural network to generate new fast-charging coefficients and update the baseline template.

[0077] (6) Differential update of local template

[0078] The updated template difference package will be encrypted and distributed to the same model of charging piles in the cloud.

[0079] The charging pile templates are merged into the old ones, the prediction engine parameters are switched, and the old templates are archived.

[0080] (7) Regional load aggregation and grid coordination

[0081] The predicted load curves of all charging piles in the area are summarized to generate the total load curve.

[0082] Converted into grid protocol commands for real-time monitoring of grid load:

[0083] If the total load in the area exceeds the safety threshold, a power reduction command will be issued to the charging pile.

[0084] The charging station adjusts its output based on the vehicle's SoC value (high SoC reduces power to 80%, low SoC reduces power to 90%).

[0085] Once the load returns to a safe range, the power limiting status is lifted.

[0086] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting the load of charging piles, characterized in that, The specific steps of this method are as follows: S100: The charging pile obtains vehicle battery management system (BMS) data in real time via the vehicle-to-pile communication protocol. Simultaneously, the cloud-based fingerprint database returns a reference charging curve template matching the battery model. This reference charging curve template includes a power function P for constant current / constant voltage stages. base (t,SoC; S200 and the dynamic prediction engine receive the baseline charging curve template and BMS data from the cloud fingerprint database in S100. Based on the real-time BMS data, they load dynamic coefficients to adjust the template parameters and calculate the predicted power P. predict (t); S300, monitor actual charging power P real (t), calculation error ΔP=|P predict -P real When ΔP > 10% and persists for 5 minutes, each charging station will encrypt and upload its error feature vector. After removing outliers in the cloud, the vectors will be grouped and aggregated by battery model to form the error feature set V for that battery model. er ; S400 uses aggregated data to train a neural network to update the curve correction model in the cloud and sends the updated template differential to the charging pile of the same model. After receiving the template, the charging pile immediately switches the prediction engine parameters and stores the old template in the historical version library. The S500 predicts the total load curve of the region output by the engine and converts it into grid communication protocol instructions. It then adjusts the power based on the actual grid load and the SoC value of the charging pile group.

2. The charging pile load prediction method according to claim 1, characterized in that, In S100, the charging pile establishes a real-time data interaction link with the vehicle battery management system (BMS) through a communication interface, sends a status read command to the BMS, and requests real-time data items including: the current remaining battery power SoC value, battery health SOH, battery model, and simultaneously collects battery temperature data T and the vehicle's unique identification code. The charging pile will standardize the collected battery model, remove redundant characters to generate a unique model code, associate it with the vehicle's unique identification code and the current charging session, bind the battery data packet to the charging pile port number and timestamp t, and send it to the pre-stored cloud fingerprint database through an encrypted transmission channel. After receiving the model code, the cloud-based fingerprint database matches it with the corresponding reference charging curve template, which includes the constant current stage power function P. base_cc (t,SoC) and the power function P during the constant voltage stage base_cv (t,SoC); The power function P during the constant current stage base_cc (t,SoC) satisfies P base_cc (t,SoC)=P rated ×f(SoC), where P rated The rated power of the charging pile is f(SoC), which is a linear function of SoC. When SoC is between 0% and 80%, f(SoC) = 1.0, and when SoC is between 80% and 95%, f(SoC) = 0.

5. The power function P during the constant voltage stage base_cv (t,SoC) satisfies P base_cv (t,SoC)=k×U rated ×I(t), where k is the conversion efficiency coefficient preset according to the battery model, with a value of 0.92-0.95, U rated Let I(t) be the rated voltage of the battery, and I(t) be the decreasing current function during the constant voltage phase. I max t0 is the maximum charging current at the start of the constant voltage phase, and t0 is the cumulative charging time from the start of the constant voltage phase. The cloud will encapsulate a benchmark charging curve template containing the power function of the constant current / constant voltage stage and send it to the corresponding charging pile. The template data includes a time axis, the power value at each time point, and a description of the function parameters.

3. The charging pile load prediction method according to claim 1, characterized in that, The dynamic prediction engine in S200 receives the benchmark charging curve template sent from the cloud fingerprint database in S100, and parses the constant current stage power function P in the template. base_cc (t,SoC) and the power function P during the constant voltage stage base_cv (t,SoC), simultaneously acquires BMS data transmitted in real time by the charging pile through the communication link, and follows P predict (t)=P base (t,SoC)×k temp (T)×k SOH (SOH) Calculate and predict the power P within 15 minutes predict (t), where k temp (T) is the temperature influence coefficient, k SOH (SOH) is the battery health impact coefficient, P base (t,SoC) is the power function for the constant current / constant voltage stage.

4. The charging pile load prediction method according to claim 3, characterized in that, In step S200, the dynamic prediction engine loads dynamic coefficient adjustment template parameters based on real-time BMS data, including: The acquired SoC value is evaluated. When the SoC is less than 30%, the constant current charging stage is initiated, and the preset fast charging coefficient k is enabled. fast That is, P predict (t)=P base (t,SoC)×k temp (T)×k SOH (SOH)×k fast For batteries with SOH ≥ 90%, k fast The value is 1.2-1.

3. For batteries with 70% ≤ SOH < 90%, k fast The value is 1.1-1.

2. For batteries with SOH < 70%, k fast Values ​​range from 1.0 to 1.1; when SoC ≥ 30%, it enters the constant current stage, k fast The value is 1.0; For the temperature influence coefficient k temp (T), when the battery temperature is in the optimal operating range of 25℃-35℃, k temp (T) takes a value of 1.

0. When the temperature is below 25℃, k increases by 1℃ for every 1℃ decrease. temp (T) decreases linearly by 0.

01. When the temperature is above 35℃, for every 1℃ increase, k temp (T) decreases linearly by 0.02, and k temp (T) The minimum value is not less than 0.8; The influence coefficient k of battery health SOH (SOH), when SOH ≥ 80%, k SOH (SOH) = 0.8 + 0.002 × (SOH - 80), where k is less than 80%. SOH (SOH) = 0.8; Based on the calculated predicted power, a power demand curve for 15 minutes is generated according to the timestamp t, and associated with the corresponding charging pile port number and vehicle unique identification code, and synchronized to the cloud.

5. The charging pile load prediction method according to claim 1, characterized in that, In S300, the charging pile collects its own actual output charging power P in real time. real (t), and associate it with the corresponding timestamp, vehicle unique identifier and charging pile port number; Call the concurrent predicted power data P generated in S200 predict (t), according to ΔP=|P predict -P real | Calculate the power error value every minute, and record the corresponding SoC value, battery temperature and SOH health status for each calculation; The error value is monitored for 5 consecutive minutes. When ΔP>10% is met for 5 consecutive minutes, the error data upload mechanism is triggered. The charging pile packages the error data that triggers the upload, forming an error feature vector V. e = [ΔP,SoC,T,SOH], where each parameter is accompanied by a corresponding timestamp and is uploaded to the cloud via an encrypted transmission channel; The cloud receives encrypted error feature vectors uploaded by each charging pile, cleans the data, removes outliers, and groups the cleaned valid error feature vectors according to the battery model obtained from S100. Error feature vectors of the same battery model are then aggregated to generate an error feature set V for that battery model. er .

6. The charging pile load prediction method according to claim 1, characterized in that, In the S500, the cloud calls a preset neural network architecture to train a curve correction model. This architecture employs three fully connected layers: an input layer, a hidden layer, and an output layer, using the generated error feature set V of the same battery model. er As a training dataset, it is divided into a training set and a validation set in an 8:2 ratio; The curve correction model is trained using the training set, and the validation set is input to complete the training. The new fast-charging coefficient k is then output. new-fast ; Based on k new-fast The power function P in the constant current stage of the reference charging curve template base_cc (t,SoC) is corrected using the formula P. new_cc (t,SoC)=P base_cc (t,SoC)×(k new-fast / k fast ), where k fast For the original fast charging coefficient in the template, a differential update package is generated. The differential update package only contains the difference data from the old template. The cloud matches the corresponding list of charging piles based on the battery model and sends the differential update package to the charging piles corresponding to the same battery model through an encrypted transmission channel. After receiving the differential update package, the charging pile automatically merges it with the old template stored locally to generate a new template, replacing the current template parameters in the dynamic prediction engine. At the same time, the old template is stored in the historical version library according to the naming rules of battery model and update timestamp, and a successful update receipt is sent to the cloud.

7. The charging pile load prediction method according to claim 1, characterized in that, In the S500, the prediction engine summarizes the energy demand curves of all charging piles in the area within 15 minutes, aligns them according to the time axis, and then uses L... total (t)=∑P predict (t) Calculate the total load curve for the region; The cloud platform converts the regional total load curve into grid communication protocol instructions, the instruction format of which includes start time, duration, required power, and maximum power ramp rate. The power grid dispatch system receives protocol instructions and monitors the actual total load value of charging piles in the area in real time. When the actual load exceeds the preset safety threshold for 30 consecutive seconds, it sends a power reduction command to the charging pile. The safety threshold is 90% of the maximum carrying capacity of the regional power grid. The power reduction command is as follows: after receiving the power reduction command, the charging pile queries the SoC value of the currently charging vehicle through the established vehicle-pile communication link. When the vehicle is charging and the SoC is ≥70%, the output power is reduced to 80% of the current actual power. When the vehicle's SoC is ≤30%, the output power is reduced to 90% of the current actual power. After the charging pile completes the power adjustment, it reports the actual power value and adjustment completion time to the power grid dispatch system. The power grid dispatch system monitors the total load of the area after the adjustment in real time. When the total load falls below the safety threshold, the power reduction state is lifted, allowing the charging pile to resume normal power output.

8. A charging pile load prediction system, applicable to the charging pile load prediction method according to any one of claims 1-7, characterized in that, The system consists of: a vehicle-pile data interaction module, a cloud fingerprint database module, a dynamic prediction engine module, a federated learning update module, and a power grid collaborative control module. The vehicle-charging pile data interaction module is used to establish a real-time communication link between the charging pile and the vehicle BMS, obtain real-time BMS data such as SoC value, battery temperature, SOH health, and battery model, and send the standardized battery model data to the cloud fingerprint library module. At the same time, it receives the benchmark charging curve template sent from the cloud. The cloud fingerprint database module pre-stores the benchmark charging curve templates corresponding to different battery models. After receiving the battery model code sent by the vehicle-charging pile data interaction module, it matches and sends out the corresponding template, and at the same time stores the historical template versions of each battery model. The dynamic prediction engine module receives templates from the cloud fingerprint library module and real-time data transmitted from the vehicle-charging data interaction module, parses the template function and determines the SoC value, activates the corresponding fast charging coefficient, and generates the predicted power and the energy demand curve within 15 minutes. The federated learning update module monitors the error between the actual charging power and the predicted power, encrypts and uploads the error feature vector to the cloud, receives the curve correction model and updated template generated by the cloud through neural network training, replaces the local template and stores the old template in the historical version library. The power grid collaborative control module summarizes the regional total load curve generated by the dynamic prediction engine module, converts it into power grid communication protocol instructions, and controls the charging pile to adjust the output power according to the vehicle SoC value.

Images