A high-efficiency charging device for a stereo garage and a control method thereof

Abstract

The present application relates to the field of charging control technology, and discloses an efficient charging device for a stereoscopic parking garage and a control method thereof, the method comprising: collecting a data set of the stereoscopic parking garage; performing intrinsic orthogonal decomposition feature extraction and constructing an initial vehicle distribution prediction model to generate a vehicle distribution prediction result; calculating a charging demand prediction result; solving based on the charging demand prediction result to obtain a partition-differentiated charging strategy; performing fault diagnosis on real-time operating data of the charging device, and dynamically adjusting the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy; based on the revised charging strategy, performing real-time control on the charging process of the charging device, and updating the initial vehicle distribution prediction model based on an incremental learning method to obtain a target vehicle distribution prediction model, thereby being able to intelligently predict charging demand, optimize charging strategy, diagnose faults in real time and dynamically adjust an efficient charging device.

Description

Technical Field

[0001] The present application relates to the field of charging control technology, and in particular to a high-efficiency charging device for a stereoscopic parking garage and a control method thereof. Background Art

[0002] With the rapid popularization of electric vehicles, stereo garages, as a parking solution for efficient use of urban space, are facing new challenges and opportunities. Traditional stereo garages mainly solve parking problems, but with the increase in the number of electric vehicles, how to provide efficient and safe charging services for a large number of electric vehicles in a limited space has become an urgent problem to be solved. At present, stereo garage charging systems generally have problems such as low charging efficiency, uneven resource allocation, and difficult load management.

[0003] The root cause of these problems lies in the lack of accurate prediction of vehicle distribution and charging demand, as well as the lack of intelligent charging strategy management. Traditional charging control methods often use fixed power allocation or simple first-come-first-served strategies, which cannot adapt to complex and changing charging demand scenarios. At the same time, due to the lack of real-time fault diagnosis and dynamic adjustment mechanisms, the reliability and safety of the system are also difficult to guarantee. Summary of the invention

[0004] The present application provides a high-efficiency charging device for a stereoscopic parking garage and a control method thereof, thereby providing a high-efficiency charging device that can intelligently predict charging needs, optimize charging strategies, diagnose faults in real time, and dynamically adjust.

[0005] In a first aspect, the present application provides a control method for a high-efficiency charging device for a stereoscopic parking garage, and the control method for a high-efficiency charging device for a stereoscopic parking garage comprises:

[0006] Collect and preprocess the vehicle distribution, charging status and grid load data of the stereo garage to obtain a preprocessed data set;

[0007] Based on the preprocessed data set, extracting features of the vehicle distribution data by intrinsic orthogonal decomposition, and constructing an initial vehicle distribution prediction model based on a long short-term memory network to generate a vehicle distribution prediction result;

[0008] According to the vehicle distribution prediction result and historical charging data, the charging probability and expected charging power of each parking space at each time point are calculated to obtain a charging demand prediction result;

[0009] Based on the charging demand prediction result, the charging efficiency maximization problem is solved to obtain a zone-differentiated charging strategy;

[0010] Analyze the real-time operation data of the charging device, perform fault diagnosis based on a mathematical model constructed based on power flow theory, and dynamically adjust the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy;

[0011] Based on the revised charging strategy, the charging process of the charging device is controlled in real time, and the initial vehicle distribution prediction model is updated based on the incremental learning method to obtain a target vehicle distribution prediction model.

[0012] A second aspect of the present application provides a high-efficiency charging device for a stereo garage, the high-efficiency charging device for a stereo garage comprising:

[0013] The acquisition module is used to collect and preprocess the vehicle distribution, charging status and power grid load data of the stereo garage to obtain a preprocessed data set;

[0014] A construction module is used to perform intrinsic orthogonal decomposition feature extraction on the vehicle distribution data based on the preprocessed data set, and to construct an initial vehicle distribution prediction model based on a long short-term memory network to generate a vehicle distribution prediction result;

[0015] A calculation module, used to calculate the charging probability and expected charging power of each parking space at each time point according to the vehicle distribution prediction result and historical charging data, so as to obtain a charging demand prediction result;

[0016] A solution module, used for solving the charging efficiency maximization problem based on the charging demand prediction result to obtain a zone-differentiated charging strategy;

[0017] An adjustment module is used to analyze the real-time operation data of the charging device, perform fault diagnosis based on a mathematical model constructed based on power flow theory, and dynamically adjust the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy;

[0018] An updating module is used to control the charging process of the charging device in real time based on the revised charging strategy, and to update the initial vehicle distribution prediction model based on an incremental learning method to obtain a target vehicle distribution prediction model.

[0019] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned control method for a high-efficiency charging device for a multi-story parking garage.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned control method for a high-efficiency charging device for a three-dimensional parking garage.

[0021] Compared with the prior art, the present application has the following beneficial effects: through the method combining intrinsic orthogonal decomposition and long short-term memory network, high-precision prediction of vehicle distribution is achieved, and based on multi-objective optimization model and mixed integer linear programming, the dual goals of maximizing charging efficiency and balancing grid load are achieved, and the utilization efficiency of charging resources is improved. Through real-time fault diagnosis and dynamic strategy adjustment, problems such as charging interface failure and cable overheating are timely identified and handled, thereby improving the safety and reliability of the system. The incremental learning method is used to update the prediction model in real time, so that the system can continuously adapt to new charging modes and demand changes, and improve the intelligent level of control. Through the partitioned differentiated charging strategy, the charging needs of different users are met, and at the same time, the stability of the charging process is guaranteed through real-time control, which improves the user's charging experience. By considering the charging strategy of grid load balance, the impact on the grid is reduced, and the coordinated operation of the charging system and the smart grid is promoted. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0024] Figure 1 It is a flow chart of a control method of a high-efficiency charging device for a stereoscopic parking garage provided by an embodiment of the present invention;

[0025] Figure 2 It is a schematic block diagram of the structure of a high-efficiency charging device for a stereoscopic parking garage provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0028] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of a control method for a high-efficiency charging device for a stereo garage in an embodiment of the present application includes:

[0030] Step 100: collecting and preprocessing vehicle distribution, charging status and grid load data of the stereo garage to obtain a preprocessed data set;

[0031] It is understandable that the execution subject of the present application may be a high-efficiency charging device for a stereoscopic parking garage, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0032] Specifically, the distribution data of vehicles is collected by vehicle detection sensors. These data usually contain timestamps, and the timestamps need to be aligned to ensure that the time series of the data is accurate and consistent. The collected vehicle distribution data is detected for outliers, and potential error data is filtered out to obtain the vehicle distribution data after preliminary processing. At the same time, the data collected by the current and voltage sensors of the charging interface are filtered to eliminate electrical noise and other signal interference to ensure the accuracy of the data. By calculating the power value of these data, valid data reflecting the current charging state, that is, charging state data, is obtained. The data collected by the power quality analyzer at the key nodes of the distribution system are Fourier transformed to decompose the various frequency components, so as to analyze the spectrum of the power system. At the same time, the grid load characteristic data is obtained through harmonic analysis. Understand the electrical harmonics existing in the power grid, which may have an important impact on the stability and efficiency of the charging system. Based on the vehicle distribution data after preliminary processing, the entry and exit time and parking position of the vehicle are statistically analyzed to form a vehicle distribution probability matrix to describe the dynamic distribution of the vehicle in the garage. The charging status data and the grid load characteristic data are correlated and analyzed. By analyzing the relationship between the charging power curve and the grid voltage fluctuation, the charging load influencing factor is obtained to help identify the specific impact of the grid status on the charging process. Based on the vehicle distribution probability matrix and the charging load influencing factor, all data are fused with multi-dimensional features and normalized to unify the scale and range of data from different sources to obtain the preprocessed data set.

[0033] Step 200: Based on the preprocessed data set, perform intrinsic orthogonal decomposition feature extraction on the vehicle distribution data, and build an initial vehicle distribution prediction model based on a long short-term memory network to generate a vehicle distribution prediction result;

[0034] Specifically, the vehicle distribution data in the preprocessed data set is reconstructed to generate a vehicle distribution snapshot matrix, in which the rows of the matrix represent the vehicle distribution at different time points, and the columns represent the occupancy status of different parking spaces. The generated vehicle distribution snapshot matrix is ​​subjected to singular value decomposition, and the original matrix is ​​decomposed into three parts: the left singular matrix, the singular value matrix, and the right singular matrix. The diagonal elements of the singular value matrix represent the importance weights of different features in the vehicle distribution data. In order to ensure the computational efficiency and generalization ability of the model, the cumulative energy contribution rate is calculated based on the singular value matrix, and the left singular vectors corresponding to the first k singular values ​​with a cumulative energy contribution rate of 95% are selected to form a feature matrix after dimensionality reduction. The feature matrix after dimensionality reduction retains the main information in the vehicle distribution data, while reducing the data dimension, improving the processing speed and performance of the model. The input layer of the LSTM is constructed, and the input layer contains k neurons, each of which corresponds to a principal component feature after dimensionality reduction. Design the hidden layer of LSTM. The hidden layer contains two LSTM layers. Each layer includes 128 LSTM units. Each unit contains a forget gate, an input gate, an output gate, and a cell state. The forget gate uses the sigmoid activation function to control the degree of information forgetting. The input gate uses the tanh activation function to generate candidate cell states. The output gate also uses the sigmoid activation function to control information output, ensuring that the LSTM network can effectively capture the long-term dependencies and dynamic change characteristics in the data when processing time series data. Construct the output layer of LSTM. The output layer consists of a fully connected layer and a softmax layer. The number of neurons in the fully connected layer is equal to the number of parking spaces. The softmax layer is responsible for normalizing the output of the fully connected layer to obtain the probability of vehicle existence in each parking space. In order to ensure the learning effect of the LSTM network, the constructed network is trained using the Adam optimizer, with a learning rate of 0.001, a batch size of 64, and 200 rounds of training. During the training process, the early stopping method is used to prevent the model from overfitting. When the performance on the validation set fails to improve for 5 consecutive rounds, the training is stopped in advance to obtain the final initial vehicle distribution prediction model. After the model training is completed, the reduced feature matrix is ​​input into the trained initial vehicle distribution prediction model for prediction. Based on the input data, the model predicts the vehicle distribution every 5 minutes in the next 24 hours and generates the vehicle distribution prediction results. To ensure the consistency of the prediction results with the actual data, the prediction results are inversely transformed from the original feature space and restored to the original high-dimensional space to obtain the final vehicle distribution prediction results.

[0035] The feature matrix after dimensionality reduction is segmented into time series to generate an input sequence of length T, where T is the time step of the long short-term memory network (LSTM). The generated input sequence is input to the input layer of the initial vehicle distribution prediction model. The model performs linear transformation on the k principal component features in each time step, calculates and generates the activation value of the input layer, which reflects the response strength of each neuron in the model. Based on the activation value of the input layer, the 128 LSTM units of the first LSTM layer perform forward propagation calculation to generate the output sequence of the first hidden layer. The function of the first LSTM layer is to preliminarily extract the time series features in the input sequence and convert the long-term dependencies in the data into information that can be further processed. The output sequence of the first hidden layer is input to the second LSTM layer, and then the time series features are extracted through 128 LSTM units. The second LSTM layer processes and refines the deep features in the input sequence to generate more accurate time series information. The output of the last time step of the output sequence of the second hidden layer is transformed by the fully connected layer. The number of neurons contained in the fully connected layer is equal to the number of parking spaces, and the output of the LSTM network is converted into interpretable vehicle distribution information. Through the calculation of the fully connected layer, the generated output will be directly associated with the distribution of parking spaces. Apply the softmax function to the output of the fully connected layer, normalize the output value to a probability distribution, and obtain the probability of vehicle presence in each parking space in the next 5-minute time interval, reflecting the possible distribution of vehicles at future time points. In order to achieve the prediction of every 5-minute interval in the next 24 hours, the above prediction process needs to be repeated 288 times. In each prediction, the previous prediction result is used as the input of the next prediction, and so on, to generate a vehicle distribution prediction probability matrix covering the next 24 hours. The prediction results are mapped from the principal component space back to the original feature space. Matrix multiplication is performed based on the vehicle distribution prediction probability matrix and the previously obtained right singular matrix. The prediction results are inversely transformed from the reduced principal component space back to the original high-dimensional space to obtain the final vehicle distribution prediction results.

[0036] Step 300: Calculate the charging probability and expected charging power of each parking space at each time point according to the vehicle distribution prediction result and historical charging data to obtain a charging demand prediction result;

[0037] It should be noted that the time series analysis of historical charging data uses the exponential smoothing method to model the average charging time and charging power of different types of vehicles in different time periods, and establishes a time-varying charging characteristic model. This model can dynamically capture the changes in charging behavior in different time periods and reflect the charging characteristics of different vehicles in different time periods. Based on the vehicle distribution prediction results and the time-varying charging characteristic model, the charging probability of each parking space in each 5-minute time interval is calculated to obtain a charging probability matrix, which reflects the possibility of charging behavior in each parking space in the future time period. Based on the charging probability matrix and the time-varying charging characteristic model, the expected charging power of each parking space in each 5-minute time interval is estimated to obtain an expected charging power matrix, which describes the charging power that each parking space may need at a specific time point. Based on the charging probability matrix and the expected charging power matrix, a weighted Markov chain model is constructed. This model describes the changes and transitional relationships of charging demand between different time periods through a state transition probability matrix. Using this model, the continuity and dynamic change law of charging demand in the time series are captured. Based on the state transition probability matrix, multiple sampling of charging demand at every 5-minute time interval in the next 24 hours is performed to generate a charging demand sample set. Through multiple sampling, a charging demand distribution covering a wide range of possibilities is obtained, which enhances the accuracy and robustness of the prediction. The generated charging demand sample set is statistically analyzed to calculate the expected value and confidence interval of the charging demand for each time interval. The statistical results not only provide a prediction of the central trend of future charging demand, but also provide a reasonable range estimate for possible demand fluctuations, and finally obtain the charging demand prediction result.

[0038] The charging probability matrix is ​​then segmented by thresholds, and parking spaces with a charging probability greater than 0.5 are marked as potential charging spaces, forming a potential charging space matrix to help identify parking spaces that are more likely to have charging behavior in the future time period and narrow the scope of subsequent calculations. Based on the time-varying charging characteristic model, the probability distribution of the vehicle type in each time interval for each potential charging space is estimated to generate a vehicle type probability distribution matrix. The probability distribution reflects the probability of different types of vehicles appearing at the potential charging space. According to the vehicle type probability distribution matrix and the charging power data in the time-varying charging characteristic model, the expectation maximization (EM) algorithm is used to iteratively calculate the expected charging power of each potential charging space to generate a preliminary expected charging power matrix. In this process, the expectation maximization algorithm gradually optimizes the estimated value of the charging power through repeated iterations, making the prediction closer to the actual situation. Based on the preliminary expected charging power matrix, a Gaussian process regression model is constructed to model the time series characteristics of the charging power and obtain the charging power prediction function. The Gaussian process regression model can well capture the trend and uncertainty of the charging power over time, and provides a continuous and smooth function description for the prediction of future charging power. Based on the charging power prediction function, the charging power at every 5-minute time interval in the next 24 hours is predicted, and the potential charging position matrix is ​​combined with the mask operation to set the prediction results that do not belong to the potential charging position to zero, and the revised expected charging power matrix is ​​obtained to ensure that the prediction results of the charging power are limited to the parking spaces where charging behavior may actually occur, avoiding misleading predictions for irrelevant sites. The revised expected charging power matrix is ​​subjected to sliding window average filtering, and the short-term fluctuations in the prediction are eliminated by local averaging of the time series, making the final expected charging power matrix smoother and more stable.

[0039] Step 400: Solve the charging efficiency maximization problem based on the charging demand prediction result to obtain a zone-differentiated charging strategy;

[0040] Specifically, the stereo garage is partitioned, and the parking spaces are divided into several charging partitions based on the floor and area information of the garage to obtain the charging partition matrix. The matrix organizes parking spaces at different locations into multiple charging partitions, so that differentiated charging strategies can be formulated for the specific needs and conditions of each partition. Based on the charging demand prediction results and the charging partition matrix, a multi-objective optimization model is constructed. The objective function of the model includes two key aspects: maximizing charging efficiency and balancing the load on the power grid. The goal of maximizing charging efficiency is to make the best use of charging resources, while the load balancing of the power grid focuses on reducing the load fluctuation on the power grid during the charging process. In addition to these goals, the model also needs to consider some important constraints, including the capacity limitation of the distribution system and the charging reliability requirements. The capacity limitation of the distribution system ensures that the charging power does not exceed the carrying capacity of the system, while the charging reliability ensures that the charging demand can be met in various possible scenarios. In order to make the optimization problem effectively solveable, the description of the above optimization problem is linearized. By converting the nonlinear constraints into piecewise linear constraints, the problem is converted into a mixed integer linear programming model. After forming a mixed integer linear programming model, the branch and bound algorithm is used to solve the charging power allocation problem and obtain the initial charging strategy. The branch and bound algorithm is a commonly used optimization algorithm that is suitable for solving complex optimization problems with discrete decision variables. During the solution process, the algorithm will gradually decompose the problem space and gradually approach the optimal solution through the process of branching and bounding to obtain a reasonable initial charging strategy. The robustness analysis of the initial charging strategy is carried out, and multiple groups of disturbance scenarios are generated through Monte Carlo simulation to evaluate the performance of the charging strategy in the face of these uncertainties. Monte Carlo simulation generates different possible scenarios through multiple random samplings to comprehensively evaluate the stability and robustness of the charging strategy. Based on the evaluation results, potential weaknesses in the strategy are identified and optimized for these weaknesses. Based on the robustness evaluation results, the initial charging strategy is adjusted and optimized. The genetic algorithm is used to fine-grainedly adjust the charging power. The genetic algorithm is an optimization algorithm based on natural selection and genetic mechanisms. By simulating the biological evolution process, it gradually optimizes the solution to obtain a better partition-differentiated charging strategy. This strategy can effectively allocate charging power among different charging zones, maximize charging efficiency, balance grid load, and ensure system robustness and reliability.

[0041] Step 500: Analyze the real-time operation data of the charging device, perform fault diagnosis based on the mathematical model constructed based on the power flow theory, and dynamically adjust the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy;

[0042] Specifically, the real-time operation data of the charging device is collected and preprocessed, including key parameters such as voltage, current and power factor. By standardizing these parameters, a standardized real-time operation data set is obtained. Based on the power flow theory, the node voltage method and branch current method are used to construct a power flow mathematical model, model the electrical characteristics of the charging system, and generate a power flow equation group of the charging system to describe the electrical relationship between the nodes of the system, including key information such as voltage amplitude, phase angle, and power flow of each branch. By solving these equations, the voltage amplitude and phase angle of each node in the system and the power flow of each branch are obtained. A fault diagnosis decision tree is constructed based on the power flow calculation results. The decision tree can effectively identify and locate common faults in the charging system, such as charging interface faults and cable overheating faults, by comparing the deviation between the power flow calculation value and the actual measurement value. The fault diagnosis results not only point out the specific fault type and location, but also provide a basis for repairing or adjusting the system operation strategy. According to the fault diagnosis results, the partition differentiated charging strategy is preliminarily corrected. The correction process includes reducing the charging power in the faulty area to avoid further equipment damage or system instability, and redistributing the charging load to ensure that the charging demand in the non-faulty area can be met, forming a preliminary revised charging strategy. Perform a safety assessment on the preliminary revised charging strategy. Verify whether the revised strategy meets the operating constraints, such as key indicators such as voltage stability and power balance, through power flow calculation. If the evaluation results show that the revised strategy does not meet these operating constraints, it is necessary to iteratively optimize the strategy and gradually adjust the charging power distribution and load scheduling until a revised charging strategy that meets all operating constraints is finally obtained.

[0043] Step 600: Based on the revised charging strategy, the charging process of the charging device is controlled in real time, and the initial vehicle distribution prediction model is updated based on the incremental learning method to obtain a target vehicle distribution prediction model.

[0044] Specifically, the revised charging strategy is parsed, the charging power curve and charging time window of each charging partition are extracted, and the charging control parameter set is obtained, which defines how the charging power should be allocated to each charging partition in different time periods to maximize the charging efficiency and ensure the stability of the system. Based on the extracted charging control parameter set, the power converter in the charging device is adjusted in real time to ensure that the charging power can be dynamically allocated according to demand. Through this process, real-time charging control instructions are generated, which directly act on the charging device to ensure that the actual charging process of each charging partition is consistent with the revised charging strategy, so as to achieve the optimal allocation and utilization of power resources. During the charging process, the actual operation data of the charging device is collected and processed in real time, including charging power, charging time and battery charge state. These actual operation data are processed to form real-time charging feedback data. Based on the real-time charging feedback data, an online learning data stream is constructed, and the features of each charging event are extracted and standardized to form an incremental learning sample set. The incremental learning sample set can reflect the changing state and demand during the charging process, which is the key to updating the vehicle distribution prediction model. The incremental learning sample set is divided into sliding windows, the window size is set to N samples, and M samples are slid forward each time to generate multiple training batches. The use of sliding windows can ensure that the model will not forget previous important information while constantly absorbing new data features, thereby improving the robustness of the model. Based on the training batches, the stochastic gradient descent algorithm is used to iteratively update the parameters of the initial vehicle distribution prediction model. The learning rate adopts an adaptive adjustment strategy during the update process to ensure that the model can quickly converge to the optimal solution while avoiding instability caused by excessive step size. While the model parameters are updated, the updated model parameters are regularized to prevent overfitting. Regularization includes L2 norm constraints and dropout operations, which can suppress the complexity of the model and enable the model to maintain good generalization ability when facing new data. The optimized model parameters are applied to the initial vehicle distribution prediction model, and the network structure and weights are further updated to finally form the target vehicle distribution prediction model.

[0045] In the embodiment of the present application, a high-precision prediction of vehicle distribution is achieved by combining the intrinsic orthogonal decomposition and the long short-term memory network. Based on the multi-objective optimization model and mixed integer linear programming, the dual goals of maximizing charging efficiency and balancing the load of the power grid are achieved, and the utilization efficiency of charging resources is improved. Through real-time fault diagnosis and dynamic strategy adjustment, problems such as charging interface failures and cable overheating are timely identified and handled, thereby improving the safety and reliability of the system. The incremental learning method is used to update the prediction model in real time, so that the system can continuously adapt to new charging modes and changes in demand, and improve the level of intelligent control. Through the partitioned differentiated charging strategy, the charging needs of different users are met, and at the same time, the stability of the charging process is guaranteed through real-time control, which improves the user's charging experience. By considering the charging strategy of the power grid load balance, the impact on the power grid is reduced, and the coordinated operation of the charging system and the smart grid is promoted.

[0046] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0047] Perform timestamp alignment and outlier detection on the data collected by the vehicle detection sensor to obtain the vehicle distribution data after preliminary processing;

[0048] Filter and calculate the power of the data collected by the current and voltage sensors of the charging interface to obtain the charging status data;

[0049] Perform Fourier transform and harmonic analysis on the data collected by the power quality analyzer at the key nodes of the distribution system to obtain the grid load characteristic data;

[0050] Based on the initially processed vehicle distribution data, statistical analysis is performed on vehicle entry and exit times and parking locations to obtain a vehicle distribution probability matrix;

[0051] According to the charging status data and grid load characteristic data, the correlation analysis between the charging power curve and the grid voltage fluctuation is carried out to obtain the charging load influencing factor;

[0052] Based on the vehicle distribution probability matrix and charging load influencing factors, the data is subjected to multi-dimensional feature fusion and normalization to obtain the preprocessed data set.

[0053] Specifically, the data collected by the vehicle detection sensor is timestamped and detected for outliers. Timestamp alignment ensures that all data can be compared and analyzed with a unified time reference. Timestamp alignment can be achieved through interpolation methods, that is, inserting missing time points or smoothing inconsistent timestamps on the time axis to ensure the temporal consistency of the data. Outliers are usually caused by sensor failure, environmental interference, or data transmission errors, which may seriously affect the accuracy of subsequent analysis. In order to detect outliers, the 3σ principle based on statistical methods is used (that is, points exceeding three times the standard deviation of the mean are considered outliers), or more complex machine learning algorithms such as isolation forests are used to identify and remove abnormal data. Through timestamp alignment and outlier detection, the vehicle distribution data after preliminary processing is obtained. The data collected by the current and voltage sensors of the charging interface are filtered. Common filtering methods include low-pass filters and Kalman filters. Low-pass filters can effectively remove high-frequency noise, while Kalman filters can achieve smooth estimation of signals in dynamic environments. The current and voltage data after filtering can be used for power calculation. The basic formula for power calculation is:

[0054] ;

[0055] in, Indicates at time The instantaneous power at and Respectively, in time The voltage and current values ​​at is the phase difference between voltage and current, is the power factor. Through this formula, the real-time power in the charging state is calculated to obtain the charging state data. The data collected by the power quality analyzer at the key nodes of the distribution system are subjected to Fourier transform and harmonic analysis to extract the characteristics of the power grid load. Fourier transform is a method of converting time domain signals into frequency domain signals. By performing Fourier transform on the power quality data, the current or voltage signal is decomposed into different frequency components, and the harmonics and noise in the system are identified. Harmonic analysis further evaluates the impact of these harmonics on the power grid load. Harmonics refer to frequency components in the current or voltage waveform that are integer multiples of the fundamental frequency. They can cause heating, vibration or failure of equipment in the power system. By identifying harmonics and evaluating their amplitude and phase, the grid load characteristic data is determined. Based on the vehicle distribution data after preliminary processing, the vehicle entry and exit time and parking position are statistically analyzed to generate a vehicle distribution probability matrix. Assuming there are N parking spaces and T time intervals, the vehicle distribution probability matrix can be expressed as a A matrix where each element Indicated in Parking space The probability of the vehicle existing at a certain point in time. This matrix can help predict the occupancy of each parking space in the parking lot at a certain point in time in the future. Based on the charging status data and the grid load characteristic data, the correlation analysis of the charging power curve and the grid voltage fluctuation is carried out to evaluate the impact of the charging behavior on the grid load and obtain the charging load impact factor. The charging load impact factor can be expressed as the correlation coefficient between two variables, such as the Pearson correlation coefficient, and its calculation formula is:

[0056] ;

[0057] in, and denote the observed values ​​of charging power and grid voltage fluctuation respectively, and are the average values ​​of these observations respectively. By calculating the correlation coefficient, the impact of charging behavior on the power grid is quantitatively evaluated to obtain the charging load impact factor. Based on the vehicle distribution probability matrix and the charging load impact factor, the data is subjected to multi-dimensional feature fusion and normalization to obtain the preprocessed data set. Multi-dimensional feature fusion refers to the comprehensive analysis of vehicle distribution data, charging status data, and power grid load characteristic data, and the extraction of representative comprehensive features through methods such as principal component analysis. Normalization is to convert these feature values ​​to the same order of magnitude to facilitate subsequent model training and prediction.

[0058] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0059] The vehicle distribution data in the preprocessed data set is reconstructed into a matrix to obtain a vehicle distribution snapshot matrix, where rows represent different time points and columns represent different parking spaces;

[0060] Perform singular value decomposition on the vehicle distribution snapshot matrix to obtain a left singular matrix, a singular value matrix, and a right singular matrix;

[0061] Based on the singular value matrix, the cumulative energy contribution rate is calculated, and the left singular vectors corresponding to the first k singular values ​​whose cumulative energy contribution rate reaches 95% are selected to obtain the characteristic matrix after dimensionality reduction;

[0062] Construct the input layer of the long short-term memory network. The input layer contains k neurons, and each neuron corresponds to a principal component feature.

[0063] Construct the hidden layer of the long short-term memory network, including two LSTM layers, each layer contains 128 LSTM units, each LSTM unit contains a forget gate, an input gate, an output gate and a cell state. The forget gate uses the sigmoid activation function to control the degree of information forgetting, the input gate uses the tanh activation function to generate candidate cell states, and the output gate uses the sigmoid activation function to control information output;

[0064] Construct the output layer of the LSTM network, which includes a fully connected layer and a softmax layer. The number of neurons in the fully connected layer is equal to the number of parking spaces. The softmax layer normalizes the output of the fully connected layer to obtain the probability of vehicle presence in each parking space.

[0065] The constructed long short-term memory network was trained using the Adam optimizer, with a learning rate of 0.001, a batch size of 64, and 200 training rounds. The early stopping method was used during the training process. When the performance on the validation set did not improve for 5 consecutive rounds, the training was stopped to obtain the initial vehicle distribution prediction model.

[0066] The reduced feature matrix is ​​input into the initial vehicle distribution prediction model to predict the vehicle distribution every 5 minutes in the next 24 hours, and the prediction result is inversely transformed with the original feature space to obtain the vehicle distribution prediction result.

[0067] Specifically, the vehicle distribution data in the preprocessed data set is reconstructed into a matrix to obtain a vehicle distribution snapshot matrix. The rows of the matrix represent different time points, while the columns represent the occupancy of different parking spaces. time points and If we have a dataset of parking spaces, then the vehicle distribution snapshot matrix is ​​a A matrix where each element in the matrix Representative in the time point, The occupancy status of parking spaces. By reconstructing the matrix of vehicle distribution data, the dynamic distribution of vehicles in the entire parking lot is expressed in a structured form. The generated vehicle distribution snapshot matrix is ​​subjected to singular value decomposition, and the matrix is ​​decomposed into three parts, namely the left singular matrix, the singular value matrix, and the right singular matrix. The mathematical form of singular value decomposition can be expressed as:

[0068] ;

[0069] in, represents the vehicle distribution snapshot matrix, is a left singular matrix, is the singular value matrix, is a right singular matrix. Each column represents a principal component of the original data, and the singular value matrix The singular values ​​in represent the weights of the corresponding principal components. The right singular matrix It represents the projection of different parking spaces on each principal component. The cumulative energy contribution rate is calculated through singular value decomposition to measure the contribution of each principal component to the original data. The calculation formula of the cumulative energy contribution rate is:

[0070] ;

[0071] in, Before The cumulative energy contribution rate of singular values ​​is is the singular value matrix The singular values, is the rank of the original data matrix. By calculating the cumulative energy contribution rate, we determine how many singular values ​​can retain more than 95% of the information of the original data. According to the left singular vectors corresponding to the singular values, we get the reduced feature matrix. Based on the feature matrix, we build an input layer of a long short-term memory network (LSTM). The input layer contains neurons, each corresponding to a principal component feature, aiming to capture the time series characteristics in the vehicle distribution data. The hidden layer of the LSTM network is constructed, including two LSTM layers, each containing 128 LSTM units. Each LSTM unit contains a forget gate, an input gate, an output gate, and a cell state. The forget gate uses a sigmoid activation function to control the degree of information forgetting, the input gate generates candidate cell states through a tanh activation function, and the output gate uses a sigmoid activation function to control information output. The LSTM network can effectively capture long-term and short-term dependencies in the data and is suitable for processing sequence data with time dependencies. In order to convert the output of the LSTM network into a probability distribution of vehicle distribution, an output layer consisting of a fully connected layer and a softmax layer is constructed. The number of neurons in the fully connected layer is equal to the number of parking spaces, and each neuron will output the occupancy status of a parking space. The softmax layer normalizes the output of the fully connected layer to generate a probability distribution, which represents the probability of the presence of a vehicle in each parking space at a specific time point. During the model training process, the Adam optimizer is used for parameter optimization. Adam is a stochastic optimization algorithm based on first-order gradients. Its characteristic is that it combines momentum and adaptive learning rate methods, so that it can converge quickly and stably. The learning rate is set to 0.001, the batch size is 64, and the training rounds are 200. In order to prevent the model from overfitting during the training process, the early stopping method is adopted, that is, when the performance on the validation set does not improve for 5 consecutive rounds, the training is stopped in advance. Through the training process, an initial vehicle distribution prediction model is obtained. After the model training is completed, the dimension-reduced feature matrix is ​​input into the initial vehicle distribution prediction model. Based on the input data, the model predicts the vehicle distribution every 5 minutes in the next 24 hours. Since the main features of vehicle distribution have been captured during the model training process, an accurate vehicle existence probability distribution can be generated. The prediction results are inversely transformed with the original feature space to restore to the original high-dimensional space to obtain the final vehicle distribution prediction results. These results can help parking lot managers understand the usage of parking spaces in advance, optimize resource allocation, and improve the operational efficiency of parking lots.

[0072] In a specific embodiment, the execution step inputs the dimension-reduced feature matrix into the initial vehicle distribution prediction model, predicts the vehicle distribution every 5 minutes in the next 24 hours, and inversely transforms the prediction result and the original feature space. The process of obtaining the vehicle distribution prediction result can specifically include the following steps:

[0073] S1: Perform time series segmentation on the reduced feature matrix to obtain an input sequence of length T, where T is the time step of the LSTM network;

[0074] S2: Input the input sequence into the input layer of the initial vehicle distribution prediction model, perform linear transformation on the k principal component features at each time step, and obtain the activation value of the input layer;

[0075] S3: Based on the activation value of the input layer, forward propagation calculation is performed through the 128 LSTM units of the first LSTM layer to obtain the output sequence of the first hidden layer;

[0076] S4: The output sequence of the first hidden layer is input into the second LSTM layer, and further time series feature extraction is performed through 128 LSTM units to obtain the output sequence of the second hidden layer;

[0077] S5: Perform a fully connected layer transformation on the output of the last time step of the output sequence of the second hidden layer to obtain neuron outputs equal to the number of parking spaces;

[0078] S6: Apply the softmax function to the output of the fully connected layer to obtain the probability of vehicle presence in each parking space within the next 5-minute time interval;

[0079] S7: Repeat steps S1 to S6 288 times in total, predicting the vehicle distribution in the next 5 minutes each time, and using the prediction results as input for the next prediction, and finally obtaining the vehicle distribution prediction probability matrix every 5 minutes in the next 24 hours;

[0080] S8: Based on the vehicle distribution prediction probability matrix and the right singular matrix, matrix multiplication operation is performed to map the prediction results from the principal component space back to the original feature space to obtain the vehicle distribution prediction results.

[0081] Specifically, the feature matrix after dimensionality reduction is segmented into time series to generate input sequences suitable for long short-term memory network processing. Each sequence corresponds to the vehicle distribution characteristics within a time window. The input sequences are input one by one into the input layer of the initial vehicle distribution prediction model. The input layer contains k neurons, each of which corresponds to a principal component feature. The k principal component features in each time step are linearly transformed to generate the activation value of the input layer. The activation value indicates the degree of response of the input feature in the neuron. The original feature is mapped to a new feature space through the weight matrix and bias. Based on the activation value of the input layer, the forward propagation calculation is performed through the 128 LSTM units of the first LSTM layer. Each LSTM unit processes the time series characteristics of the input data through a complex gating mechanism, including a forget gate, an input gate, and an output gate. The forget gate uses the sigmoid activation function to determine which information needs to be discarded. The input gate generates candidate cell states through the tanh activation function and updates the network's memory in combination with the existing cell states. The output gate determines which information needs to be output through the sigmoid function. After processing by these units, the network can effectively capture the long-term and short-term dependencies of vehicle distribution and obtain the output sequence of the first hidden layer. The output sequence of the first hidden layer is input to the second LSTM layer to extract deep temporal features. The second LSTM layer also contains 128 LSTM units, which can capture more complex time series patterns by further processing the output of the first hidden layer. The output sequence of the second hidden layer is generated through layer-by-layer extraction of the two LSTM layers. The output of the last time step of the output sequence of the second hidden layer is transformed by the fully connected layer. The number of neurons in the fully connected layer is equal to the number of parking spaces, and the output of this layer will be directly associated with the occupancy status of each parking space. The role of the fully connected layer is to convert the temporal features extracted by the LSTM layer into a specific probability distribution of vehicle existence. The softmax function is applied to the output of the fully connected layer. The softmax function converts the output value of the fully connected layer into a probability distribution, so that the probability of vehicle existence in each parking space in the next 5-minute time interval can be clearly expressed. Specifically, the formula of the softmax function is:

[0082] ;

[0083] in, Indicates The probability that there is a vehicle in a parking space, It is the fully connected layer The output of a neuron, is the total number of parking spaces. Through this process, the vehicle distribution probability of each parking space in the next 5 minutes is obtained. In order to predict the vehicle distribution every 5 minutes in the next 24 hours, steps S1 to S6 are repeated 288 times. The prediction result of each time is used as the input for the next prediction, so as to gradually calculate the vehicle distribution probability matrix for the entire 24 hours. Through successive iterations, the LSTM network can use past prediction information to continuously adjust and optimize the next prediction results to form a set of continuous prediction sequences. Based on the generated vehicle distribution prediction probability matrix and the previously obtained right singular matrix, matrix multiplication is performed to map the prediction results from the principal component space back to the original feature space. The low-dimensional features after dimensionality reduction are converted back into high-dimensional data corresponding to the actual vehicle distribution so that these prediction results can be intuitively understood and used. The specific mapping process can be expressed as:

[0084] ;

[0085] in, is the vehicle distribution matrix mapped back from the principal component space, is the previous left singular matrix, is a diagonal matrix containing the predictions, is the transpose of a right singular matrix.

[0086] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0087] Perform time series analysis on historical charging data, and model the average charging time and charging power of different time periods and different types of vehicles based on the exponential smoothing method to obtain a time-varying charging characteristic model;

[0088] Based on the vehicle distribution prediction results and the time-varying charging characteristic model, the charging probability of each parking space in each 5-minute time interval is calculated to obtain the charging probability matrix;

[0089] Based on the charging probability matrix and the time-varying charging characteristic model, the expected charging power of each parking space in each 5-minute time interval is estimated to obtain the expected charging power matrix;

[0090] Based on the charging probability matrix and the expected charging power matrix, a weighted Markov chain model is constructed to model the timing dependence of charging demand and obtain the state transition probability matrix.

[0091] Based on the state transition probability matrix, multiple samples are taken of the charging demand at 5-minute intervals in the next 24 hours to obtain a charging demand sample set.

[0092] Statistical analysis is performed on the charging demand sample set, and the expected value and confidence interval of the charging demand at each time interval are calculated to obtain the charging demand prediction result.

[0093] Specifically, a time series analysis of historical charging data is performed to capture the patterns of charging behavior over time. Historical charging data usually includes the charging duration and charging power of different types of vehicles in different time periods. In order to accurately model these changing characteristics, the exponential smoothing method is used to model the trend and seasonal components in the data. The exponential smoothing method assigns different weights to historical data, with recent data having a larger weight and long-term data having a smaller weight, thereby making a more accurate prediction of future charging behavior. The mathematical expression of the exponential smoothing method is:

[0094] ;

[0095] in, Indicates at a point in time The smoothed value of It's time point The actual observed value of is the smoothing coefficient, which ranges from 0 to 1 and controls the relative weight of recent data and long-term data in the model. The value of is used to obtain the smooth curve that is most suitable for modeling the charging time and charging power of the vehicle, and generate a time-varying charging characteristic model. Based on the vehicle distribution prediction results and the generated time-varying charging characteristic model, the charging probability of each parking space in each 5-minute time interval is calculated to form a charging probability matrix. Assuming that the vehicle distribution prediction result is the probability of the existence of a vehicle in each parking space at a certain point in the future, then by combining the time-varying charging characteristic model, the probability that the vehicle needs to be charged at that point in time is derived. For example, assuming that the probability of the existence of a vehicle in a parking space at a specific point in time is 0.8, and according to the time-varying charging characteristic model, the average charging demand probability of this type of vehicle during this period is 0.5, then the probability that the parking space needs to be charged at this point in time is 0.4. By calculating all parking spaces and time points, the charging probability matrix is ​​obtained. Based on the charging probability matrix and the time-varying charging characteristic model, the expected charging power of each parking space in each 5-minute time interval is estimated to form an expected charging power matrix. The charging probability is converted into a specific power demand to more accurately predict future power loads. For example, if the charging probability of a parking space at a certain point in time is 0.4, and the corresponding time-varying charging characteristic model shows that the average charging power of this type of vehicle is 20kW, then the expected charging power of the parking space at that point in time is 8kW. Through this calculation method, the expected charging power matrix can describe in detail the power requirements of each parking space at each point in the future. In order to model the temporal dependence of charging demand, a weighted Markov chain model is constructed based on the charging probability matrix and the expected charging power matrix. The Markov chain model is a statistical model that describes the change of system state through state transition probability. In this model, each state represents the charging demand at a point in time, and the transition between states is determined by the charging probability matrix and the expected charging power matrix. The state transition probability matrix can be expressed by the following formula:

[0096] ;

[0097] in, Indicates from the state Transfer to state The probability of Indicates time point state. These transition probabilities are affected by the charging probability and expected power. For example, if the charging probability is high and the expected power is large at the current time point, the probability of maintaining a high charging demand at the next time point will also be correspondingly large. The weighted Markov chain model is used to describe the dynamic changes of charging demand in the time series. Based on the state transition probability matrix, the charging demand at every 5-minute time interval in the next 24 hours is sampled multiple times to obtain a charging demand sample set. Through methods such as Monte Carlo simulation, a large number of possible charging demand scenarios are generated to reflect the uncertainty and volatility of charging demand. By analyzing these samples, the robustness of the charging strategy is evaluated and more options are provided for the dispatch of the power system. The charging demand sample set is statistically analyzed, and the expected value and confidence interval of the charging demand at each time interval are calculated to obtain the final charging demand forecast result. The expected value represents the average level of charging demand at a certain point in time, while the confidence interval represents the range of possible fluctuations in charging demand under a certain confidence level.

[0098] In a specific embodiment, the execution step estimates the expected charging power of each parking space in each 5-minute time interval based on the charging probability matrix and the time-varying charging characteristic model, and the process of obtaining the expected charging power matrix may specifically include the following steps:

[0099] Perform threshold segmentation on the charging probability matrix, mark the parking spaces with a charging probability greater than 0.5 as potential charging spaces, and obtain a potential charging space matrix;

[0100] Based on the time-varying charging characteristic model, the probability distribution of the vehicle type at each potential charging position in each time interval is estimated to obtain the vehicle type probability distribution matrix.

[0101] According to the vehicle type probability distribution matrix and the charging power data in the time-varying charging characteristic model, the expected charging power of each potential charging position is iteratively calculated using the expectation maximization algorithm to obtain a preliminary expected charging power matrix.

[0102] Based on the preliminary expected charging power matrix, a Gaussian process regression model is constructed to model the time series characteristics of the charging power and obtain the charging power prediction function.

[0103] Based on the charging power prediction function, the charging power at every 5-minute time interval in the next 24 hours is predicted, and the mask operation is performed in combination with the potential charging position matrix to obtain the revised expected charging power matrix;

[0104] The corrected expected charging power matrix is ​​subjected to sliding window average filtering to obtain the expected charging power matrix.

[0105] Specifically, the charging probability matrix is ​​segmented by thresholds to identify the parking spaces that are most likely to require charging in the future time period. The charging probability matrix is ​​denoted as ,in Indicates parking space. represents the time interval, and represents the charging probability of each parking space in each time interval. Set a threshold , usually set to 0.5, is used to determine whether a parking space is a potential charging space. For each parking space, if its charging probability Greater than , then mark it as a potential charging position to obtain the potential charging position matrix

[0106] ;

[0107] Based on the time-varying charging characteristic model, the probability distribution of the vehicle type at each potential charging position in each time interval is estimated. The vehicle type is set to , and assume that the vehicle type distribution of each potential charging position in each time interval conforms to a certain probability distribution, recorded as . This probability reflects the possibility of a certain type of vehicle appearing at a potential charging position at a specific time point, forming a vehicle type probability distribution matrix. After obtaining the vehicle type probability distribution matrix, the expected charging power of each potential charging position is calculated in combination with the charging power data in the time-varying charging characteristic model. The expected charging power of each potential charging position is optimized by iterative calculation using the expectation maximization algorithm. ,in Representative parking space In time interval The steps of the expectation maximization algorithm can be expressed as: Step E (expectation step): Calculate the expected value of the charging power given the current parameters .

[0108] ;

[0109] in, Is the vehicle type The average charging power. Step M (maximization step): maximize the log-likelihood function and update the estimated value of the charging power. Through multiple iterations, the algorithm gradually converges to a local optimal solution and finally obtains the preliminary expected charging power matrix. Based on the preliminary expected charging power matrix, a Gaussian process regression model (GPR) is constructed to model the time series characteristics of charging power. GPR sets a Gaussian process as the prior distribution, and then updates this distribution in combination with the observed data to obtain the posterior distribution and form a charging power prediction function. ,in Indicates time:

[0110] ;

[0111] in, is the mean function, usually set to 0, is the kernel function, which defines different time points and One of the commonly used kernel functions is the square exponential kernel function:

[0112] ;

[0113] in, represents the variance of the signal, is a scale parameter that controls time correlation. Through the training and application of the GPR model, the charging power prediction function is obtained , and use this to predict the charging power every 5 minutes in the next 24 hours. To ensure the accuracy of the prediction results, combined with the potential charging position matrix Perform a mask operation, that is, set the predicted power to zero for parking spaces that are not potential charging spaces:

[0114] ;

[0115] in, is the expected charging power matrix corrected after the mask operation. In order to smooth the prediction results, the corrected expected charging power matrix Perform sliding window average filtering. The sliding window average filtering process can be expressed as follows:

[0116] ;

[0117] in, Indicates the time interval The smooth charging power value, is the window size. Through the processing of sliding window, the prediction noise caused by short-term fluctuations is eliminated, making the final expected charging power matrix smoother and more stable.

[0118] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0119] The three-dimensional parking garage is partitioned, and the parking spaces are divided into several charging partitions based on the floor and area information to obtain a charging partition matrix;

[0120] Based on the charging demand prediction results and the charging partition matrix, a multi-objective optimization model is constructed. The objective function includes maximizing charging efficiency and balancing the grid load. The constraints include the capacity limitation of the distribution system and the charging reliability requirements. The optimization problem description is obtained.

[0121] The optimization problem description is linearized, and the nonlinear constraints are transformed into piecewise linear constraints to obtain a mixed integer linear programming model;

[0122] Based on the mixed integer linear programming model, the branch and bound algorithm is used to solve the charging power allocation and obtain the initial charging strategy;

[0123] Conduct robustness analysis on the initial charging strategy, generate multiple disturbance scenarios through Monte Carlo simulation, evaluate the performance of the charging strategy under uncertainty, and obtain robustness evaluation results;

[0124] Based on the robustness evaluation results, the initial charging strategy is adjusted and optimized, and the charging power is fine-grainedly adjusted using a genetic algorithm to obtain a partition-differentiated charging strategy.

[0125] Specifically, the garage is partitioned into several charging partitions. Each parking space is classified based on the floor and area information. Assuming that the garage has multiple floors and each floor is divided into several areas, each area can be regarded as an independent charging partition. Through this partitioning process, a charging partition matrix is ​​generated, and the elements of the matrix represent the charging partition to which each parking space belongs. Assuming there is Parking spaces and charging partitions, the charging partition matrix can be expressed as a The matrix, the elements in the matrix are binary, if the parking space Belongs to the charging zone , then the corresponding matrix element is 1, otherwise it is 0. Based on the charging demand prediction results and the charging partition matrix, a multi-objective optimization model is constructed. The objective function of this model includes two main aspects: maximizing charging efficiency and balancing the grid load. The goal of maximizing charging efficiency is to ensure that the charging needs of vehicles are met as much as possible under limited resources; grid load balancing focuses on how to avoid excessive load fluctuations on the grid while meeting the charging needs. It is described by the following objective function:

[0126] ;

[0127] ;

[0128] in, Indicates the charging efficiency, Indicates grid load fluctuations, is the charging efficiency coefficient, Indicates at time Parking In the partition The charging power, is the total charging power, is the total number of time intervals. The constraints of the model include the capacity limit of the distribution system and the charging reliability requirement. The capacity limit of the distribution system ensures that the total charging power will not exceed the carrying capacity of the distribution system, which is expressed as:

[0129] ;

[0130] in, is the maximum capacity of the distribution system. The charging reliability requirement ensures that the charging demand can be met with a certain probability. Since some constraints in the model may be nonlinear, in order to simplify the solution process, the nonlinear constraints are converted into piecewise linear constraints to obtain a mixed integer linear programming (MILP) model. The linearization method includes using piecewise linear functions to approximate the original nonlinear function, or decomposing the nonlinear constraints into a combination of multiple linear constraints. After obtaining the mixed integer linear programming model, the branch and bound algorithm is used to solve the charging power allocation. The branch and bound algorithm is a commonly used discrete optimization algorithm that recursively divides the problem space and gradually approaches the optimal solution. The basic idea of ​​the algorithm is to decompose the problem into several sub-problems, and then use the pruning strategy to exclude those branches that cannot contain the optimal solution, and finally converge to the global optimal solution. The robustness analysis of the initial charging strategy is performed. Multiple sets of disturbance scenarios are generated through Monte Carlo simulation to evaluate the performance of the charging strategy under uncertainty. Monte Carlo simulation is a method that generates a large number of possible scenarios through random sampling, which can simulate a variety of possible situations of charging demand and grid load fluctuations. For each simulation scenario, the performance indicators of the charging strategy, such as charging efficiency and grid load fluctuations, are calculated to obtain robustness evaluation results to help identify potential weaknesses in the strategy. Based on the robustness evaluation results, the initial charging strategy is adjusted and optimized. A genetic algorithm is used to perform fine-grained adjustments to the charging power. A genetic algorithm is an optimization algorithm based on natural selection and genetic mechanisms that gradually optimizes solutions by simulating the biological evolution process. The basic steps of a genetic algorithm include selection, crossover, mutation, and survival competition. Through multiple generations of evolution, the algorithm gradually converges to a global optimal or suboptimal solution, and obtains a partition-differentiated charging strategy.

[0131] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0132] Collect and preprocess the real-time operation data of the charging device, including voltage, current, and power factor parameters, to obtain a standardized real-time operation data set;

[0133] A mathematical model of power flow is constructed based on the node voltage method and the branch current method, the electrical characteristics of the charging system are modeled, and the power flow equations of the charging system are obtained;

[0134] Solve the power flow equations to obtain the voltage amplitude and phase angle of each node in the system and the power flow of each branch;

[0135] Based on the power flow calculation results, a fault diagnosis decision tree is constructed. By comparing the deviation between the calculated value and the measured value, the charging interface fault and the cable overheating fault are identified and located to obtain the fault diagnosis result.

[0136] According to the fault diagnosis results, the zone-differentiated charging strategy is modified, including reducing the charging power in the fault area and redistributing the charging load, to obtain a preliminary modified charging strategy;

[0137] A safety assessment is conducted on the initially revised charging strategy, and power flow calculation is used to verify whether the revised strategy meets the operating constraints. If not, it is iteratively optimized to ultimately obtain the revised charging strategy.

[0138] Specifically, the real-time operation data of the charging device are comprehensively collected and preprocessed. These data usually include key parameters such as voltage, current and power factor. These data are standardized to ensure that they can be compared on the same scale. The standardization process usually uses z-score standardization to convert the data of each parameter into a standard normal distribution with a mean of 0 and a standard deviation of 1 to obtain a standardized real-time operation data set. A mathematical model of the power flow of the charging system is constructed based on the node voltage method and the branch current method. The node voltage method and the branch current method are two basic methods in power system analysis. The node voltage method is mainly used to solve the voltage of each node in the system, while the branch current method is used to calculate the current in each branch. Through these two methods, a group of power flow equations for the charging system is established. The group of power flow equations describes the electrical relationship between each node in the system. For the node voltage method, the power balance equation can be expressed as:

[0139] ;

[0140] in, and Respectively represent The active power and reactive power of each node, Is a node The voltage, Is a node and nodes The admittance between represents conjugation, is the total number of nodes in the system. This set of equations is nonlinear and requires an iterative solution method, such as the Newton-Raphson method, to solve the voltage amplitude and phase angle of each node in the system, as well as the power flow of each branch. A fault diagnosis decision tree is constructed based on the power flow calculation results. The fault diagnosis decision tree is a tool commonly used for fault detection and location. By inputting the deviation between the calculated value and the measured value into the decision tree, common faults in the charging system, such as charging interface faults and cable overheating faults, are identified. Assume there is a node The calculated voltage value is The measured voltage is , define a deviation

[0141] ;

[0142] if If a certain threshold is exceeded, it may mean that there is a fault in the node. Similarly, a deviation in the power flow in the cable It can also be used to detect cable overheating or other types of faults. Through the structure of the decision tree, it is automatically determined whether these deviations are beyond the normal range, and the fault type and location are identified to obtain the fault diagnosis results. According to the fault diagnosis results, the charging strategy is corrected to reduce the charging power in the fault area and redistribute the charging load to ensure the safe and stable operation of the system. For example, if the charging interface in a certain area fails, the revised strategy may reduce the charging power in that area and transfer part of the load to other unaffected areas. The preliminary revised charging strategy can effectively reduce the impact of the fault on the entire system and prevent the spread of the fault by adjusting the load distribution. After the preliminary revised strategy is formulated, a safety assessment is performed. The core of the safety assessment is to verify whether the revised strategy meets the operating constraints of the system. In order to conduct a safety assessment, the power flow calculation is performed again to confirm whether the revised charging power distribution is within the carrying capacity of the distribution system and to check whether key operating conditions such as voltage stability are met. If the power flow calculation results show that the revised strategy fails to meet these constraints, further iterative optimization is required. The iterative optimization process may involve adjusting the load distribution, recalculating the power flow, and even modifying the priority of the fault area until a final revised strategy that meets all operating constraints is found. For example, in some cases, it may be necessary to reduce the impact of a fault on the system by adding backup power or adjusting the load in other areas. In each iteration, the system adjusts the strategy based on the latest power flow calculation results until all safety evaluation indicators meet the requirements.

[0143] In a specific embodiment, the process of executing step 600 may specifically include the following steps:

[0144] Analyze the revised charging strategy, extract the charging power curve and charging time window of each charging partition, and obtain the charging control parameter set;

[0145] Based on the charging control parameter set, the power converter of the charging device is adjusted in real time to achieve dynamic allocation of charging power and obtain real-time charging control instructions;

[0146] Collect and process the actual operating data of the charging device, including charging power, charging time and battery charge status, to obtain real-time charging feedback data;

[0147] Based on real-time charging feedback data, an online learning data stream is constructed to extract and standardize the features of each charging event to obtain an incremental learning sample set.

[0148] The incremental learning sample set is divided into sliding windows, the window size is set to N samples, and M samples are slid forward each time to obtain multiple training batches;

[0149] Based on the training batch, the stochastic gradient descent algorithm is used to iteratively update the parameters of the initial vehicle distribution prediction model, and the learning rate adopts an adaptive adjustment strategy to obtain the updated model parameters;

[0150] Regularize the updated model parameters, including L2 norm constraints and dropout operations, to prevent overfitting and obtain optimized model parameters;

[0151] The optimized model parameters are applied to the initial vehicle distribution prediction model, the network structure and weights are updated, and the target vehicle distribution prediction model is obtained.

[0152] Specifically, the revised charging strategy is parsed to extract the charging power curve and charging time window of each charging partition, and describe the charging demand and load of each partition in a specific time period. The charging power curve can be represented by a time series, recording the charging power demand of a partition at different time points, and the charging time window defines the distribution of these power demands on the time axis. By parsing the revised charging strategy, a charging control parameter set is obtained, which includes the charging power demand of each partition, the corresponding time window, and other control parameters such as load priority. Based on the charging control parameter set, the power converter of the charging device is adjusted in real time to achieve dynamic allocation of charging power. The power converter controls the actual charging power received by each partition by adjusting the output of the converter. According to the data in the charging control parameter set, the charging control instruction is generated in real time to guide the operation of the power converter. For example, assuming that the charging demand of a partition is high during the peak period, the power converter will be instructed to allocate more power resources to the partition during this period, and reduce the power output of the partition during other periods to achieve the balance and optimization of the charging power. During the charging process, the operating status of the charging device is monitored in real time. The actual operating data of the charging device is collected and processed. These data include information such as charging power, charging time, and battery state of charge. By processing these data, real-time charging feedback data is obtained to reflect the actual situation of the current charging process. By comparing it with the previously formulated charging strategy, possible deviations in the actual implementation process are found. Based on the real-time charging feedback data, an online learning data stream is constructed to improve and optimize the charging strategy. Online learning is a machine learning method that continuously updates the model to adapt to new data and is suitable for real-time control in a dynamic environment. The features of each charging event are extracted, such as the charging start time, duration, and shape of the power curve, and these features are standardized to obtain an incremental learning sample set. Through standardization, the data in the incremental learning sample set is converted into a standard normal distribution with a mean of 0 and a variance of 1, which is conducive to model training and convergence. In order to improve the performance of the model, the incremental learning sample set is divided into sliding windows. The window size is set to N samples, and M samples are slid forward each time to generate multiple training batches. The sliding window method ensures that the model can use the latest data each time it is updated, while retaining some historical information to avoid the model from being overly dependent on data at a single time point. Assuming the window size is 100 samples and the sliding step size is 10 samples, the latest 100 sample data will be used each time the model is updated, but only 10 samples are updated each time compared to the previous training batch, so as to achieve a smooth transition. After completing the data preparation, the parameters of the initial vehicle distribution prediction model are iteratively updated based on the training batch using the stochastic gradient descent algorithm. The stochastic gradient descent algorithm is a commonly used optimization algorithm suitable for model training of large-scale data sets.For each training batch, the algorithm calculates the gradient of the loss function and updates the model parameters in the direction of gradient descent. The learning rate η is a key parameter in the stochastic gradient descent algorithm, which controls the step size of each parameter update. In order to improve the convergence speed and avoid falling into the local optimum, an adaptive adjustment strategy is adopted to dynamically adjust the learning rate according to the changes in the loss function:

[0153] ;

[0154] in, It is The learning rate at update time, is the decay rate. With adaptive learning rate, the model can learn quickly at the beginning of training, and then gradually reduce the step size to adjust the parameters more finely. During the process of updating the model parameters, the updated model parameters are regularized to prevent overfitting. Overfitting can cause the model to perform well on the training set but poorly on new data. For this reason, commonly used regularization techniques include L2 norm constraints and dropout operations. The L2 norm constraint limits the size of the model parameters by adding a weight penalty term to the loss function:

[0155] ;

[0156] in, is the loss function after regularization, is the original loss function, are model parameters, is the regularization coefficient. The dropout operation prevents the model from overfitting by randomly discarding some neurons and improves the generalization ability of the model. After the regularization process is completed, the optimized model parameters will be applied to the initial vehicle distribution prediction model to update the network structure and weights and obtain the target vehicle distribution prediction model.

[0157] The control method of the efficient charging device for the stereo garage in the embodiment of the present application is described above. The efficient charging device 10 for the stereo garage in the embodiment of the present application is described below. Figure 2 In the embodiment of the present application, an efficient charging device 10 for a stereo garage includes:

[0158] The acquisition module 11 is used to collect and preprocess the vehicle distribution, charging status and power grid load data of the stereo garage to obtain a preprocessed data set;

[0159] A construction module 12 is used to extract the intrinsic orthogonal decomposition features of the vehicle distribution data based on the preprocessed data set, and to construct an initial vehicle distribution prediction model based on a long short-term memory network to generate a vehicle distribution prediction result;

[0160] The calculation module 13 is used to calculate the charging probability and expected charging power of each parking space at each time point according to the vehicle distribution prediction result and the historical charging data, and obtain the charging demand prediction result;

[0161] A solution module 14 is used to solve the charging efficiency maximization problem based on the charging demand prediction result to obtain a zone-differentiated charging strategy;

[0162] The adjustment module 15 is used to analyze the real-time operation data of the charging device, perform fault diagnosis based on the mathematical model constructed by the power flow theory, and dynamically adjust the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy;

[0163] The updating module 16 is used to control the charging process of the charging device in real time based on the revised charging strategy, and to update the initial vehicle distribution prediction model based on the incremental learning method to obtain the target vehicle distribution prediction model.

[0164] Through the collaboration of the above components, the method of combining intrinsic orthogonal decomposition and long short-term memory network is used to achieve high-precision prediction of vehicle distribution. Based on multi-objective optimization model and mixed integer linear programming, the dual goals of maximizing charging efficiency and balancing grid load are achieved, and the utilization efficiency of charging resources is improved. Through real-time fault diagnosis and dynamic strategy adjustment, problems such as charging interface failure and cable overheating are timely identified and handled, which improves the safety and reliability of the system. The incremental learning method is used to update the prediction model in real time, so that the system can continuously adapt to new charging modes and demand changes, and improve the intelligent level of control. Through the partitioned differentiated charging strategy, the charging needs of different users are met, and the stability of the charging process is guaranteed through real-time control, which improves the user's charging experience. Through the charging strategy that considers the grid load balance, the impact on the grid is reduced, and the coordinated operation of the charging system and the smart grid is promoted.

[0165] The present application also provides an electronic device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the control method for a high-efficiency charging device for a multi-story parking garage in the above-mentioned embodiments.

[0166] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the control method for a high-efficiency charging device for a multi-story parking garage.

[0167] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0168] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0169] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A control method for an efficient charging device for a stereo garage, It is characterized in that The method comprises: The vehicle distribution, charging status and grid load data of the stereo garage are collected and preprocessed to obtain a preprocessed data set; specifically, the data collected by the vehicle detection sensor are timestamp aligned and outlier detected to obtain the vehicle distribution data after preliminary processing; the data collected by the current and voltage sensors of the charging interface are filtered and the power is calculated to obtain the charging status data; the data collected by the power quality analyzer of the key nodes of the distribution system are subjected to Fourier transform and harmonic wave analysis to obtain the grid load characteristic data; based on the vehicle distribution data after preliminary processing, the vehicle entry and exit time and parking position are statistically analyzed to obtain the vehicle distribution probability matrix; according to the charging status data and the grid load characteristic data, the charging power curve and the grid voltage fluctuation are correlated and analyzed to obtain the charging load influencing factor; based on the vehicle distribution probability matrix and the charging load influencing factor, the data is subjected to multi-dimensional feature fusion and normalization to obtain the preprocessed data set; Based on the preprocessed data set, extracting features of the vehicle distribution data by intrinsic orthogonal decomposition, and constructing an initial vehicle distribution prediction model based on a long short-term memory network to generate a vehicle distribution prediction result; According to the vehicle distribution prediction result and historical charging data, the charging probability and expected charging power of each parking space at each time point are calculated to obtain a charging demand prediction result; Based on the charging demand prediction result, the charging efficiency maximization problem is solved to obtain a zone-differentiated charging strategy; Analyze the real-time operation data of the charging device, perform fault diagnosis based on a mathematical model constructed based on power flow theory, and dynamically adjust the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy; Based on the revised charging strategy, the charging process of the charging device is controlled in real time, and the initial vehicle distribution prediction model is updated based on the incremental learning method to obtain a target vehicle distribution prediction model.

2. The control method of the high-efficiency charging device for a stereo garage according to claim 1, It is characterized in that Based on the preprocessed data set, intrinsic orthogonal decomposition feature extraction is performed on the vehicle distribution data, and an initial vehicle distribution prediction model is constructed based on a long short-term memory network to generate a vehicle distribution prediction result, including: Reconstructing the vehicle distribution data in the preprocessed data set into a matrix to obtain a vehicle distribution snapshot matrix, wherein rows represent different time points and columns represent different parking spaces; Performing singular value decomposition on the vehicle distribution snapshot matrix to obtain a left singular matrix, a singular value matrix, and a right singular matrix; Based on the singular value matrix, the cumulative energy contribution rate is calculated, and the left singular vectors corresponding to the first k singular values ​​whose cumulative energy contribution rate reaches 95% are selected to obtain the characteristic matrix after dimensionality reduction; Constructing an input layer of a long short-term memory network, wherein the input layer comprises k neurons, each neuron corresponding to a principal component feature; Constructing a hidden layer of a long short-term memory network, including two LSTM layers, each layer containing 128 LSTM units, each LSTM unit containing a forget gate, an input gate, an output gate and a cell state, the forget gate uses a sigmoid activation function to control the degree of information forgetting, the input gate uses a tanh activation function to generate candidate cell states, and the output gate uses a sigmoid activation function to control information output; Constructing an output layer of the long short-term memory network, including a fully connected layer and a softmax layer, wherein the number of neurons in the fully connected layer is equal to the number of parking spaces, and the softmax layer normalizes the output of the fully connected layer to obtain the probability of vehicle existence in each parking space; The constructed long short-term memory network was trained using the Adam optimizer, with a learning rate of 0.001, a batch size of 64, and 200 training rounds. The early stopping method was used during the training process. When the performance on the validation set did not improve for 5 consecutive rounds, the training was stopped to obtain the initial vehicle distribution prediction model. The reduced-dimensional feature matrix is ​​input into the initial vehicle distribution prediction model to predict the vehicle distribution at every 5-minute interval in the next 24 hours, and the prediction result is inversely transformed with the original feature space to obtain the vehicle distribution prediction result.

3. The control method of the high-efficiency charging device for a stereo garage according to claim 2, It is characterized in that The reduced dimension feature matrix is ​​input into the initial vehicle distribution prediction model to predict the vehicle distribution every 5 minutes in the next 24 hours, and the prediction result is inversely transformed with the original feature space to obtain the vehicle distribution prediction result, including: S1: performing time series segmentation on the feature matrix after dimension reduction to obtain an input sequence of length T, where T is the time step of the LSTM network; S2: Input the input sequence into the input layer of the initial vehicle distribution prediction model, perform linear transformation on the k principal component features of each time step, and obtain the activation value of the input layer; S3: Based on the activation value of the input layer, forward propagation calculation is performed through 128 LSTM units of the first LSTM layer to obtain the output sequence of the first hidden layer; S4: Input the output sequence of the first hidden layer into the second LSTM layer, perform further time series feature extraction through 128 LSTM units, and obtain the output sequence of the second hidden layer; S5: performing a fully connected layer transformation on the output of the last time step of the output sequence of the second hidden layer to obtain neuron outputs equal to the number of parking spaces; S6: Apply a softmax function to the output of the fully connected layer to obtain the probability of a vehicle existing in each parking space within the next 5-minute time interval; S7: Repeat steps S1 to S6 288 times in total, predicting the vehicle distribution in the next 5 minutes each time, and using the prediction results as input for the next prediction, and finally obtaining the vehicle distribution prediction probability matrix every 5 minutes in the next 24 hours; S8: Based on the vehicle distribution prediction probability matrix and the right singular matrix, matrix multiplication operation is performed to map the prediction result from the principal component space back to the original feature space to obtain the vehicle distribution prediction result.

4. The control method of the high-efficiency charging device for a stereo garage according to claim 1, It is characterized in that The charging probability and expected charging power of each parking space at each time point are calculated according to the vehicle distribution prediction result and the historical charging data to obtain the charging demand prediction result, including: Performing time series analysis on the historical charging data, modeling the average charging time and charging power of vehicles of different types in different time periods based on the exponential smoothing method, and obtaining a time-varying charging characteristic model; Based on the vehicle distribution prediction result and the time-varying charging characteristic model, the charging probability of each parking space in each 5-minute time interval is calculated to obtain a charging probability matrix; Based on the charging probability matrix and the time-varying charging characteristic model, an expected charging power of each parking space in each 5-minute time interval is estimated to obtain an expected charging power matrix; Based on the charging probability matrix and the expected charging power matrix, a weighted Markov chain model is constructed to model the timing dependency of the charging demand and obtain a state transition probability matrix; Based on the state transition probability matrix, multiple sampling of charging demand at every 5-minute time interval in the next 24 hours is performed to obtain a charging demand sample set; Statistical analysis is performed on the charging demand sample set, and the expected value and confidence interval of the charging demand at each time interval are calculated to obtain a charging demand prediction result.

5. The control method of the high-efficiency charging device for a stereo garage according to claim 4, It is characterized in that The method of estimating the expected charging power of each parking space in each 5-minute time interval based on the charging probability matrix and the time-varying charging characteristic model to obtain an expected charging power matrix includes: Performing threshold segmentation on the charging probability matrix, marking parking spaces with a charging probability greater than 0.5 as potential charging spaces, and obtaining a potential charging space matrix; Based on the time-varying charging characteristic model, a probability distribution estimate is performed on the vehicle type of each potential charging position in each time interval to obtain a vehicle type probability distribution matrix; According to the vehicle type probability distribution matrix and the charging power data in the time-varying charging characteristic model, an expectation maximization algorithm is used to iteratively calculate the expected charging power of each potential charging position to obtain a preliminary expected charging power matrix; Based on the preliminary expected charging power matrix, a Gaussian process regression model is constructed to model the time series characteristics of the charging power to obtain a charging power prediction function; Based on the charging power prediction function, the charging power at every 5-minute time interval in the next 24 hours is predicted, and a mask operation is performed in combination with the potential charging position matrix to obtain a revised expected charging power matrix; The corrected expected charging power matrix is ​​subjected to sliding window average filtering to obtain the expected charging power matrix.

6. The control method of the high-efficiency charging device for a stereo garage according to claim 1, It is characterized in that The method solves the charging efficiency maximization problem based on the charging demand prediction result to obtain a zone-differentiated charging strategy, including: The three-dimensional parking garage is partitioned, and the parking spaces are divided into several charging partitions based on the floor and area information to obtain a charging partition matrix; Based on the charging demand prediction result and the charging partition matrix, a multi-objective optimization model is constructed, wherein the objective function includes maximizing charging efficiency and balancing the grid load, and the constraint conditions include the capacity limitation of the distribution system and the charging reliability requirement, thereby obtaining a description of the optimization problem; Linearizing the optimization problem description, converting the nonlinear constraints into piecewise linear constraints, and obtaining a mixed integer linear programming model; Based on the mixed integer linear programming model, a branch and bound algorithm is used to solve the charging power allocation to obtain an initial charging strategy; Performing a robustness analysis on the initial charging strategy, generating multiple groups of disturbance scenarios through Monte Carlo simulation, evaluating the performance of the charging strategy under uncertainty, and obtaining a robustness evaluation result; Based on the robustness evaluation results, the initial charging strategy is adjusted and optimized, and the charging power is fine-grainedly adjusted using a genetic algorithm to obtain a partition-differentiated charging strategy.

7. The control method of the high-efficiency charging device for a stereo garage according to claim 1, It is characterized in that The real-time operation data of the charging device is analyzed, a mathematical model constructed based on power flow theory is used for fault diagnosis, and the partition-differentiated charging strategy is dynamically adjusted according to the diagnosis result to obtain a revised charging strategy, including: Collect and preprocess the real-time operation data of the charging device, including voltage, current, and power factor parameters, to obtain a standardized real-time operation data set; A power flow mathematical model is constructed based on the node voltage method and the branch current method to model the electrical characteristics of the charging system and obtain the power flow equations of the charging system; Solving the power flow equations to obtain the voltage amplitude and phase angle of each node in the system and the power flow of each branch; Based on the power flow calculation result, a fault diagnosis decision tree is constructed, and by comparing the deviation between the calculated value and the measured value, the charging interface fault and the cable overheating fault are identified and located to obtain a fault diagnosis result; According to the fault diagnosis result, the zone-differentiated charging strategy is modified, including reducing the charging power of the fault area and redistributing the charging load, to obtain a preliminary modified charging strategy; A safety assessment is performed on the initially revised charging strategy, and power flow calculation is used to verify whether the revised strategy meets the operation constraints. If not, it is iteratively optimized to ultimately obtain the revised charging strategy.

8. The control method of the high-efficiency charging device for a stereo garage according to claim 1, It is characterized in that The charging process of the charging device is controlled in real time based on the revised charging strategy, and the initial vehicle distribution prediction model is updated based on the incremental learning method to obtain a target vehicle distribution prediction model, including: Analyzing the modified charging strategy, extracting the charging power curve and charging time window of each charging partition, and obtaining a charging control parameter set; Based on the charging control parameter set, a power converter of the charging device is adjusted in real time to achieve dynamic allocation of charging power and obtain a real-time charging control instruction; Collect and process the actual operating data of the charging device, including charging power, charging time and battery charge status, to obtain real-time charging feedback data; Based on the real-time charging feedback data, an online learning data stream is constructed, and the features of each charging event are extracted and standardized to obtain an incremental learning sample set; Divide the incremental learning sample set into sliding windows, set the window size to N samples, and slide forward M samples each time to obtain multiple training batches; Based on the training batch, the parameters of the initial vehicle distribution prediction model are iteratively updated using a stochastic gradient descent algorithm, and the learning rate adopts an adaptive adjustment strategy to obtain updated model parameters; Regularize the updated model parameters, including L2 norm constraints and dropout operations, to prevent overfitting and obtain optimized model parameters; The optimized model parameters are applied to the initial vehicle distribution prediction model, the network structure and weights are updated, and the target vehicle distribution prediction model is obtained.

9. An efficient charging device for a stereo garage, It is characterized in that A control method for executing a high-efficiency charging device for a stereoscopic parking garage according to any one of claims 1 to 8, wherein the high-efficiency charging device for a stereoscopic parking garage comprises: The acquisition module is used to collect and preprocess the vehicle distribution, charging status and grid load data of the stereo garage to obtain a preprocessed data set; specifically including: performing timestamp alignment and outlier detection on the data collected by the vehicle detection sensor to obtain the vehicle distribution data after preliminary processing; performing filtering processing and power calculation on the data collected by the current and voltage sensors of the charging interface to obtain the charging status data; performing Fourier transform and harmonic wave analysis on the data collected by the power quality analyzer at the key nodes of the distribution system to obtain the grid load characteristic data; based on the vehicle distribution data after preliminary processing, statistical analysis is performed on the vehicle entry and exit time and parking position to obtain the vehicle distribution probability matrix; according to the charging status data and the grid load characteristic data, correlation analysis is performed on the charging power curve and the grid voltage fluctuation to obtain the charging load influencing factor; based on the vehicle distribution probability matrix and the charging load influencing factor, multi-dimensional feature fusion and normalization processing are performed on the data to obtain the preprocessed data set; A construction module is used to perform intrinsic orthogonal decomposition feature extraction on the vehicle distribution data based on the preprocessed data set, and to construct an initial vehicle distribution prediction model based on a long short-term memory network to generate a vehicle distribution prediction result; A calculation module, used to calculate the charging probability and expected charging power of each parking space at each time point according to the vehicle distribution prediction result and historical charging data, so as to obtain a charging demand prediction result; A solution module, used for solving the charging efficiency maximization problem based on the charging demand prediction result to obtain a zone-differentiated charging strategy; An adjustment module is used to analyze the real-time operation data of the charging device, perform fault diagnosis based on a mathematical model constructed based on power flow theory, and dynamically adjust the partition-differentiated charging strategy according to the diagnosis result to obtain a revised charging strategy; An updating module is used to control the charging process of the charging device in real time based on the revised charging strategy, and to update the initial vehicle distribution prediction model based on an incremental learning method to obtain a target vehicle distribution prediction model.

Images