Charging pile use frequency analysis method based on charging frequency data

Abstract

The invention relates to the field of charging pile use, and discloses a charging pile use frequency analysis method based on charging frequency data, which comprises the following steps: acquiring charging frequency original data of a plurality of charging piles at different time periods and different dates, performing structured processing in combination with a time label, and constructing a charging behavior data set containing a time dimension label; introducing a sliding window algorithm and a periodic spectrum analysis method, and performing frequency feature extraction on the charging behavior data set; using a density peak clustering algorithm to classify the use frequency matrixes of different charging piles under the same time scale, and identifying use preferences and active features of the charging piles under various typical time scenes; based on a lightweight time sequence neural network structure, a frequency fluctuation sensing mechanism is introduced; and predicting the use condition of the charging pile in a future time period based on the trained frequency model in combination with regression prediction and an anomaly detection algorithm. The method has the advantage of improving operation scheduling.

Description

Technical Field

[0001] This invention relates to the field of charging pile usage, specifically a method for analyzing the usage frequency of charging piles based on charging frequency data. Background Technology

[0002] With the rapid development of new energy vehicles, charging piles, as an important component of their supporting infrastructure, are constantly increasing in number. To improve the operational efficiency and user experience of charging piles, operators typically use charging frequency data to analyze the usage frequency of charging piles, thereby optimizing site selection, capacity planning, and maintenance scheduling. Common analysis methods include counting the number of times each charging pile is used within a certain time window and using this to assess its activity and utilization rate. However, existing technologies for frequency analysis generally have a specific problem: they fail to fully consider the fluctuations in charging frequency over time. For example, some charging piles may be frequently used on weekdays but idle on weekends, or experience concentrated peaks in specific seasons. Most current methods only perform static statistics based on the total number of uses, ignoring the changing patterns in the time series, which may lead to distorted analysis results and fail to accurately reflect the actual usage patterns of charging piles. Therefore, it is necessary to design a charging pile usage frequency analysis method based on charging frequency data to improve operational scheduling. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] To address the shortcomings of existing technologies, this invention provides a method for analyzing the usage frequency of charging piles based on charging frequency data, which has the advantage of improving operation scheduling and solves the problems mentioned in the background technology.

[0005] (II) Technical Solution

[0006] To achieve the aforementioned goal of improving operational scheduling, this invention provides the following technical solution: a method for analyzing the usage frequency of charging piles based on charging frequency data, comprising the following steps:

[0007] We acquire raw data on the number of times each charging station is charged at different times and on different dates, and then perform structured processing by combining time labels to construct a charging behavior dataset with time dimension annotations.

[0008] By introducing the sliding window algorithm and periodic spectrum analysis method, frequency features are extracted from the charging behavior dataset, and a multi-scale charging pile usage frequency feature matrix is ​​constructed.

[0009] Using the density peak clustering algorithm, the usage frequency matrix of different charging piles under the same time scale is classified to identify the usage preferences and activity characteristics of charging piles in various typical time scenarios.

[0010] Based on a lightweight temporal neural network structure, a frequency fluctuation sensing mechanism is introduced to achieve adaptive training of the charging pile usage frequency model, and to monitor the model's response capability to abnormal periods in real time and automatically adjust the weight parameters.

[0011] Based on the trained frequency model, combined with regression prediction and anomaly detection algorithms, the usage of charging piles in the future is predicted, and a frequency utilization report is generated.

[0012] Preferably, the process of constructing a charging behavior dataset with time dimension annotation is as follows:

[0013] Obtain raw charging record data from multiple charging stations;

[0014] The timestamp information in the charging record is standardized, and the time is discretized according to the set time granularity;

[0015] The discretized time information is used as a time label and mapped to the number of charging times to generate a structured charging behavior data table.

[0016] By incorporating auxiliary time stamp information, a charging behavior dataset is ultimately formed.

[0017] Preferably, the process of introducing the sliding window algorithm and periodic spectrum analysis method is as follows:

[0018] Set the time length and step size parameters of the sliding window, and slice the charging behavior dataset along the time axis to form a window data subset with temporal continuity;

[0019] For the charging frequency sequence data within each time window, Fourier transform is used to extract the periodic signal components of the charging frequency sequence data;

[0020] Based on the frequency domain signal characteristics, the main frequency components and amplitude information are identified, and a frequency feature vector corresponding to each time window is formed.

[0021] The frequency feature vectors of all time windows are aggregated to generate a basic frequency feature set.

[0022] Preferably, the frequency feature extraction process for the charging behavior dataset is as follows:

[0023] After extracting the basic frequency feature set, we analyze the changing trends of each frequency component in different time windows to identify the dominant frequency.

[0024] The intensity of the dominant frequency on different charging piles is normalized to measure the differences in the spatial distribution of frequency characteristics.

[0025] A frequency intensity sorting mechanism is used to extract the high-frequency and low-frequency terms of the frequency response characteristics of each charging pile, and a frequency feature template with a unified format is constructed.

[0026] The frequency behavior of charging piles is characterized by frequency templates.

[0027] Preferably, the process of constructing the multi-scale charging pile usage frequency feature matrix is ​​as follows:

[0028] Multiple time scales are set to extend the frequency feature template over time.

[0029] At each time scale, the frequency characteristics of different charging piles at that scale are summarized to form a corresponding frequency sub-matrix;

[0030] Matrix standardization techniques are used to unify frequency feature matrices at different scales to the same numerical range.

[0031] By splicing the frequency sub-matrices of each time scale according to the time hierarchy, a unified multi-scale charging pile usage frequency feature matrix is ​​constructed.

[0032] Preferably, the process of classifying the usage frequency matrix of different charging piles under the same time scale is as follows:

[0033] For the constructed multi-scale frequency feature matrix, cluster analysis is performed on each time scale dimension.

[0034] The density peak clustering algorithm is used to identify the center point and divide the clusters based on the local density and distance of frequency feature points;

[0035] The silhouette coefficient of each cluster is evaluated, and the time scale with the best clustering effect is selected as the typical classification scale.

[0036] Output the category label of each charging station at a typical scale.

[0037] Preferably, the process of identifying the usage preferences and activity characteristics of charging piles in various typical time scenarios is as follows:

[0038] The average, peak, and fluctuation range of charging frequency of charging piles in each cluster at different time periods were statistically analyzed.

[0039] By combining external environmental labels, we can compare and analyze the response differences of various types of charging piles in typical scenarios.

[0040] Based on the correspondence between statistical indicators and scene tags, the activity curves and preferred scene tags of charging piles in various typical time scenarios are extracted.

[0041] Based on the activity curve, a feature vector for charging pile scenario adaptation is constructed.

[0042] Preferably, the adaptive training process for the charging pile usage frequency model is as follows:

[0043] The charging pile scenario adaptation feature vector is used as input and imported into a lightweight temporal neural network structure;

[0044] A frequency fluctuation sensing mechanism is introduced, and the sensitivity parameters of the model to abrupt frequency changes are set.

[0045] A multi-stage warm-up training strategy was adopted to gradually optimize the time memory units in the network structure;

[0046] The model's prediction accuracy in typical scenarios is dynamically monitored during training.

[0047] Preferably, the automatic adjustment process of the weight parameters is as follows:

[0048] During training, the gradient update rate and loss function changes of each hidden layer are recorded in real time.

[0049] By incorporating an attention mechanism, dynamic weights are assigned to input data within different time windows;

[0050] For periods of high frequency fluctuation, the parameter weights during model training are automatically increased.

[0051] Preferably, the process for predicting the usage of charging stations in the future is as follows:

[0052] Based on the frequency prediction model that has been trained, input the frequency feature vector and scene label of the latest time node;

[0053] The system makes progressive predictions for multiple future time windows and outputs the predicted frequency of charging piles within each window.

[0054] By combining the prediction results with the anomaly detection model, the time periods with future usage peaks or troughs can be identified;

[0055] By comparing the predicted frequency results with the maximum service capacity of charging piles, a future usage pressure map is generated, and a frequency utilization report is output.

[0056] (III) Beneficial Effects

[0057] Compared with existing technologies, this invention provides a method for analyzing the usage frequency of charging piles based on charging frequency data, which has the following beneficial effects:

[0058] This invention, by introducing a sliding window algorithm and periodic spectrum analysis, can fully extract the potential periodicity and temporal dynamic features from charging frequency data, constructing a frequency feature matrix with multi-scale temporal resolution. Utilizing density peak clustering to classify frequency behavior helps reveal the preference patterns and operational activity of different types of charging piles in specific time scenarios, providing theoretical support for differentiated operation and maintenance and load optimization. Introducing a frequency fluctuation sensing mechanism and a lightweight temporal neural network structure enables the model to possess good adaptive and anomaly response capabilities, dynamically adjusting the learning weights for sudden high-frequency or low-frequency usage behavior, thereby improving prediction accuracy and generalization ability. Finally, by fusing regression prediction and anomaly detection mechanisms, it can accurately predict charging pile usage trends in future time periods and generate frequency utilization maps and reports, providing data-driven decision-making basis for power dispatching, charging network planning, and resource optimization, significantly improving the operational efficiency and service quality of charging infrastructure. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the method of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] Please see Figure 1 As shown in the figure, the charging pile usage frequency analysis method based on charging frequency data according to an embodiment of the present invention includes the following steps:

[0063] S1: Obtain raw data on the number of times multiple charging stations charge at different times and on different dates, combine it with time labels for structured processing, and construct a charging behavior dataset with time dimension annotation.

[0064] The process of constructing the charging behavior dataset with time dimension annotation in S1 is as follows:

[0065] Obtain raw charging record data from multiple charging stations, including charging station number, start charging time, end charging time, charging duration, and number of charging sessions;

[0066] The timestamp information in the charging record is standardized, and the time is discretized according to the set time granularity;

[0067] Discretized time information is used as time labels and mapped to charging frequency data to generate a structured charging behavior data table. Under the two-dimensional coordinates of each charging pile number and its time granularity label, the total number of charging times within the corresponding time period is calculated. A two-dimensional structured table is constructed: the charging pile number is used as the row primary key, the time label as the column primary key, and the corresponding cell content is the charging frequency value for that time period. If a charging pile has no charging record within a certain time period, it is filled with 0 to avoid matrix sparsity affecting subsequent analysis. The time span is uniformly processed to ensure that the time labels of all charging piles are consistent within the analysis period, guaranteeing the alignment of the time series analysis. The output is in a standard structured data format and serves as the input dataset for the next stage of label enhancement processing.

[0068] Further, auxiliary time label information such as holidays, weather, and urban traffic flow is introduced to enhance the sensitivity of the dataset to changes in time scenarios, ultimately forming a charging behavior dataset with multi-dimensional time characteristics.

[0069] S2: Introducing the sliding window algorithm and periodic spectrum analysis method, frequency features are extracted from the charging behavior dataset to construct a multi-scale charging pile usage frequency feature matrix.

[0070] The process of introducing the sliding window algorithm and periodic spectrum analysis method in S2 is as follows:

[0071] The time length and step size parameters of the sliding window are set, and the charging behavior dataset is sliced ​​along the time axis to form a window data subset with temporal continuity. Based on the discretized time tags and charging frequency data, time series data at the single charging pile or site dimension is constructed, in the form: S=[(t1,c1),(t2,c2),...,(t n ,c n )], where t i Let ci be the time label, and ci be the number of charges corresponding to that moment. A sliding window length W and a step size S are set to divide the time-continuous subsequence. Two common settings are as follows: Fixed window + fixed step size: suitable for uniformly sampled data streams; Dynamic window: the window size can be dynamically adjusted according to holidays, peak periods, etc., and the sequence is sliced ​​in chronological order from the starting moment using a sliding window to generate multiple window data subsets Win1, Win2,...,Win k Each window contains a sequence of consecutive charging counts: Win i =[c i ,c i+1 ,...,c i+ W -1 Adjacent windows slide according to a step size S, allowing for partial overlap;

[0072] For the charging frequency sequence data within each time window, Fourier transform is used to extract the periodic signal components of the charging frequency sequence data; the window data is normalized to eliminate the noise influence caused by baseline differences between different windows; and the discrete time series within the window is processed. i Applying the Fast Fourier Transform (FFT), the time-domain signal is converted into a frequency-domain signal: F i (k)=∑_{n=0}^{W-1}Win i (n)*e^(-j2πkn / W), where F i (k) represents the complex amplitude of the k-th frequency component, and W is the window length. The amplitude of each frequency component is extracted as: |F i (k)|=sqrt(Re(F i (k)) 2 +Im(F i (k)) 2 The frequency index k is used to measure the energy intensity of the frequency component in the original signal and to map the discrete frequency index k in the Fourier domain back to the actual frequency, which facilitates the subsequent interpretation of features.

[0073] Based on the frequency domain signal characteristics, the main frequency components and amplitude information are identified, and a frequency feature vector corresponding to each time window is formed.

[0074] The frequency feature vectors of all time windows are aggregated to generate a basic frequency feature set, which serves as the input basis for subsequent multi-scale modeling.

[0075] The frequency feature extraction process for the charging behavior dataset in S2 is as follows:

[0076] After extracting the basic frequency feature set, the changing trends of each frequency component in different time windows are analyzed to identify the dominant frequencies with statistical significance. A two-dimensional matrix F(t,f) is constructed, where t represents different time windows; f represents the preset frequency component; and the cell value is the amplitude of the frequency in that time window. The average amplitude and variance of each frequency are calculated across all time windows. Methods such as moving average or exponential smoothing are used to analyze the time-varying curve of each frequency. Analysis of variance or nonparametric tests are performed on the frequency amplitude sequence. It is determined whether the frequency components have stability or periodic intensity fluctuations on the time axis. The frequencies that pass the significance test are selected as the dominant frequencies.

[0077] The intensity of the dominant frequency on different charging piles is normalized to measure the differences in the spatial distribution of frequency characteristics.

[0078] A frequency intensity sorting mechanism is adopted to extract the high-frequency and low-frequency terms of the frequency response characteristics of each charging pile, and construct a frequency feature template with a unified format. The frequency feature vector of each charging pile is sorted from high to low amplitude. Threshold parameters are defined to select the strongest high-frequency term and the weakest low-frequency term respectively. A template vector with a unified format is constructed. If the actual frequency count of a certain pile is insufficient, zero padding or mean padding can be used to ensure the uniformity of the template structure. Frequency units or period values ​​are retained to facilitate feature readability.

[0079] The frequency behavior of charging piles is characterized by frequency templates, providing structured data support for constructing a frequency feature matrix.

[0080] The process of constructing the multi-scale charging pile usage frequency feature matrix in S2 is as follows:

[0081] Multiple time scales, such as hourly, daily, and weekly, are set to expand the frequency feature templates. Multiple time scales are defined, such as: hourly (e.g., 1 hour, 3 hours); daily (e.g., 24 hours, day / night); weekly (e.g., weekday / weekend, every 7 days). Appropriate granularity is set based on actual business cycles, grid operating frequency, or user charging behavior. A corresponding time scale label is added to each charging behavior record. Each frequency feature template is bound to its corresponding time scale for subsequent layered processing. Charging frequency sequences are recalculated or summarized within different time scales to update the frequency templates. For hourly levels, a sliding window is constructed by hour; for daily / weekly levels, data from a full day or week is aggregated before extracting frequency components.

[0082] At each time scale, the frequency characteristics of different charging piles at that scale are summarized to form a corresponding frequency sub-matrix; for each time scale S i Extract the frequency response templates of all charging piles within the corresponding time range; form FreqMatrix_S i That is, the frequency submatrix at this time scale;

[0083] Matrix normalization techniques are used to unify the frequency feature matrices at different scales to the same numerical range, which facilitates subsequent model processing.

[0084] The frequency sub-matrices of each time scale are concatenated according to the time hierarchy to construct a unified multi-scale charging pile usage frequency feature matrix for subsequent classification and modeling. Using charging piles as row units, their frequency features at different time scales are concatenated horizontally. Each row corresponds to one charging pile; each column represents the dominant frequency and its intensity value for that pile at different time scales. The total number of columns is: frequency dimension × number of time scales × number of frequency template elements. The charging pile usage frequency matrix is ​​used as input for multi-scale clustering analysis, periodic behavior prediction models, and abnormal usage pattern recognition applications.

[0085] Example 2

[0086] like Figure 1 As shown, a method for analyzing the usage frequency of charging piles based on charging frequency data also includes the following steps:

[0087] S3: Using the density peak clustering algorithm, the usage frequency matrix of different charging piles at the same time scale is classified to identify the usage preferences and activity characteristics of charging piles in various typical time scenarios.

[0088] The process of classifying the usage frequency matrix of different charging piles at the same time scale in S3 is as follows:

[0089] For the constructed multi-scale frequency feature matrix, cluster analysis is performed on each time scale dimension.

[0090] The density peak clustering algorithm is used to identify center points and divide clusters based on the local density and distance of frequency feature points; the local density is calculated for each frequency feature point using the following formula:

[0091]

[0092] In the formula, d ij Let d be the distance between point i and point j. c The cutoff distance;

[0093] For each point, calculate the minimum distance δ to points with higher density;

[0094] Sort by ρ*δ, select the first few points as cluster centers; other points are assigned to the cluster of the nearest, higher-density point.

[0095] For each cluster, the silhouette coefficient is evaluated, and the time scale with the best clustering effect is selected as the typical classification scale; for each sample i, the silhouette coefficient s(i) is calculated, as follows:

[0096]

[0097] In the formula, a(i) is the average distance between the sample and other samples in the same cluster, and b(i) is the average distance between the sample and the samples in the nearest other cluster; the average silhouette coefficients at different time scales are compared, and the time scale with the highest value is selected; the time scale is defined as the typical frequency behavior classification scale.

[0098] Output the category label of each charging pile at a typical scale to provide a category basis for subsequent scenario preference analysis.

[0099] The process of identifying the usage preferences and activity characteristics of charging piles in various typical time scenarios in S3 is as follows:

[0100] The average, peak, and fluctuation range of charging frequency of charging piles in each cluster at different time periods were statistically analyzed.

[0101] By combining external environmental labels, we can compare and analyze the response differences of various types of charging piles in typical scenarios. External time label information includes, but is not limited to: whether it is a holiday; weather type (sunny / rainy / snowy, etc.); urban traffic flow level (high / medium / low). We associate time slices T_i with corresponding labels to form a {T_i:label} structure. Within multiple time periods with the same environmental label (such as holidays, rainy days): we compare the changes in charging frequency in different categories; we extract sensitivity differences: such as the frequency of one type increases on holidays, while the frequency of another type decreases.

[0102] Based on the correspondence between statistical indicators and scene tags, the activity curves and preference scene tags of charging piles in various typical time scenarios are extracted; the frequency indicators of each cluster in each time slice are connected to form an activity curve; the higher the activity, the larger the peak of the curve; multiple curves are generated according to different dimensions; for time points / environment combinations with high activity, their corresponding tags are extracted; these are used as typical usage scenarios of this type of charging pile preference.

[0103] Based on the activity curve, a charging pile scenario adaptation feature vector is constructed for model adaptive training.

[0104] S4: Based on a lightweight temporal neural network structure, a frequency fluctuation sensing mechanism is introduced to realize adaptive training of the charging pile usage frequency model, and to monitor the model's response capability to abnormal periods in real time and automatically adjust the weight parameters.

[0105] The adaptive training process for the charging pile usage frequency model in S4 is as follows:

[0106] The charging pile scenario adaptation feature vector is used as input and imported into a lightweight temporal neural network structure;

[0107] A frequency fluctuation sensing mechanism is introduced, and sensitivity parameters for the model to mutation frequency changes are set; mutation frequency segments are identified in the training data; fluctuation level or rate of change index is labeled for each input sequence; sensitivity parameters are set to control the mutation response intensity and are dynamically updated during training, so that the model gradually adapts to high-frequency or outlier changes.

[0108] A multi-stage warm-up training strategy is adopted to progressively optimize the temporal memory units in the network structure to adapt to changes in multi-scale temporal features. Stage 1: Short-term temporal feature training: The model is trained using high-frequency (small-granularity) input data; focusing on short-term behavior prediction, the parameters of the temporal memory units are initialized. Stage 2: Mid-term cycle adaptation training (weekly): Data with longer time windows is introduced; the model is trained to identify mid-term trends and enhance the length of the memory chain. Stage 3: Long-term stability enhancement training (monthly or seasonal): Multi-scale data is fully input; the model weights are comprehensively optimized to improve long-term prediction stability.

[0109] During training, the model's prediction accuracy in typical scenarios is dynamically monitored to achieve adaptive fitting capability for different time scenarios.

[0110] The automatic adjustment process of the weight parameters in S4 is as follows:

[0111] During training, the gradient update rate and loss function changes of each hidden layer are recorded in real time.

[0112] By incorporating an attention mechanism, dynamic weights are assigned to input data within different time windows;

[0113] For periods of high frequency fluctuation, the parameter weights during model training are automatically increased to enhance the model's ability to adapt to abnormal situations.

[0114] Gradient normalization and weight regularization mechanisms are introduced to prevent overfitting and maintain the stability of the model training process.

[0115] S5: Based on the trained frequency model, combined with regression prediction and anomaly detection algorithms, the usage of charging piles in the future time period is predicted, and a frequency utilization report is generated.

[0116] The process of predicting the usage of charging piles in the future time period in S5 is as follows:

[0117] Based on the already trained frequency prediction model, the system takes the frequency feature vector and scene label of the latest time point as input; extracts the frequency feature vector corresponding to the current time point in real time from the data acquisition system; simultaneously acquires external scene label information of the current time point, such as holiday identification, weather conditions, traffic flow, etc., to form a scene label vector; standardizes the input feature vector to ensure uniform input scale; concatenates or fuses the frequency feature vector and scene label vector to construct a joint input vector; and inputs the joint vector into the trained frequency prediction neural network model as the initial state for prediction.

[0118] The system progressively predicts the frequency of charging piles within multiple future time windows, outputting the predicted frequency value for each window. It sets a set of future time windows for prediction, with each window corresponding to a certain time length. Utilizing the model's recursive structure or sliding window mechanism, it sequentially generates the predicted charging frequency values ​​for each future time window based on the prediction results of the previous window and the corresponding scene label. For each prediction step, it inputs the current window's predicted value and scene information, outputting the frequency prediction for the next window. The prediction results include multi-scale frequency features, ensuring comprehensive prediction across the time dimension.

[0119] By combining the prediction results with the anomaly detection model, the time periods with future usage peaks or troughs are identified; the predicted frequency sequence is input into the preset anomaly detection model, and by comparing the predicted frequency with the historical normal range, the time windows with predicted frequencies significantly higher or lower are identified, generating a list of peak / trough time periods for scheduling and management reference.

[0120] By comparing the predicted frequency results with the maximum service capacity of charging piles, a future usage pressure map is generated, and a frequency utilization report is output.

[0121] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0122] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing the usage frequency of charging piles based on charging frequency data, characterized in that, Includes the following steps: We acquire raw data on the number of times each charging station is charged at different times and on different dates, and then perform structured processing by combining time labels to construct a charging behavior dataset with time dimension annotations. By introducing the sliding window algorithm and periodic spectrum analysis method, frequency features are extracted from the charging behavior dataset, and a multi-scale charging pile usage frequency feature matrix is ​​constructed. Using the density peak clustering algorithm, the usage frequency matrix of different charging piles under the same time scale is classified to identify the usage preferences and activity characteristics of charging piles in various typical time scenarios. Based on a lightweight temporal neural network structure, a frequency fluctuation sensing mechanism is introduced to achieve adaptive training of the charging pile usage frequency model, and to monitor the model's response capability to abnormal periods in real time and automatically adjust the weight parameters. Based on the trained frequency model, combined with regression prediction and anomaly detection algorithms, the usage of charging piles in the future is predicted, and a frequency utilization report is generated.

2. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 1, characterized in that, The process of constructing a charging behavior dataset with time dimension annotation is as follows: Obtain raw charging record data from multiple charging stations; The timestamp information in the charging record is standardized, and the time is discretized according to the set time granularity; The discretized time information is used as a time label and mapped to the number of charging times to generate a structured charging behavior data table. By incorporating auxiliary time stamp information, a charging behavior dataset is ultimately formed.

3. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 2, characterized in that, The process of introducing the sliding window algorithm and periodic spectrum analysis method is as follows: Set the time length and step size parameters of the sliding window, and slice the charging behavior dataset along the time axis to form a window data subset with temporal continuity; For the charging frequency sequence data within each time window, Fourier transform is used to extract the periodic signal components of the charging frequency sequence data; Based on the frequency domain signal characteristics, the main frequency components and amplitude information are identified, and a frequency feature vector corresponding to each time window is formed. The frequency feature vectors of all time windows are aggregated to generate a basic frequency feature set.

4. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 3, characterized in that, The process of extracting frequency features from the charging behavior dataset is as follows: After extracting the basic frequency feature set, we analyze the changing trends of each frequency component in different time windows to identify the dominant frequency. The intensity of the dominant frequency on different charging piles is normalized to measure the differences in the spatial distribution of frequency characteristics. A frequency intensity sorting mechanism is used to extract the high-frequency and low-frequency terms of the frequency response characteristics of each charging pile, and a frequency feature template with a unified format is constructed. The frequency behavior of charging piles is characterized by frequency templates.

5. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 4, characterized in that, The process of constructing the multi-scale charging pile usage frequency feature matrix is ​​as follows: Multiple time scales are set to extend the frequency feature template over time. At each time scale, the frequency characteristics of different charging piles at that scale are summarized to form a corresponding frequency sub-matrix; Matrix standardization techniques are used to unify frequency feature matrices at different scales to the same numerical range. By splicing the frequency sub-matrices of each time scale according to the time hierarchy, a unified multi-scale charging pile usage frequency feature matrix is ​​constructed.

6. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 5, characterized in that, The process of classifying the usage frequency matrix of different charging piles on the same time scale is as follows: For the constructed multi-scale frequency feature matrix, cluster analysis is performed on each time scale dimension. The density peak clustering algorithm is used to identify the center point and divide the clusters based on the local density and distance of frequency feature points; The silhouette coefficient of each cluster is evaluated, and the time scale with the best clustering effect is selected as the typical classification scale. Output the category label of each charging station at a typical scale.

7. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 6, characterized in that, The process of identifying the usage preferences and activity characteristics of charging stations in various typical time scenarios is as follows: The average, peak, and fluctuation range of charging frequency of charging piles in each cluster at different time periods were statistically analyzed. By combining external environmental labels, we can compare and analyze the response differences of various types of charging piles in typical scenarios. Based on the correspondence between statistical indicators and scene tags, the activity curves and preferred scene tags of charging piles in various typical time scenarios are extracted. Based on the activity curve, a feature vector for charging pile scenario adaptation is constructed.

8. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 7, characterized in that, The adaptive training process for the charging pile usage frequency model is as follows: The charging pile scenario adaptation feature vector is used as input and imported into a lightweight temporal neural network structure; A frequency fluctuation sensing mechanism is introduced, and the sensitivity parameters of the model to abrupt frequency changes are set. A multi-stage warm-up training strategy was adopted to gradually optimize the time memory units in the network structure; The model's prediction accuracy in typical scenarios is dynamically monitored during training.

9. The method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 8, characterized in that, The process of automatically adjusting the weight parameters is as follows: During training, the gradient update rate and loss function changes of each hidden layer are recorded in real time. By incorporating an attention mechanism, dynamic weights are assigned to input data within different time windows; For periods of high frequency fluctuation, the parameter weights during model training are automatically increased.

10. A method for analyzing the usage frequency of charging piles based on charging frequency data according to claim 9, characterized in that, The process of predicting the usage of charging stations in the future is as follows: Based on the frequency prediction model that has been trained, input the frequency feature vector and scene label of the latest time node; The system makes progressive predictions for multiple future time windows and outputs the predicted frequency of charging piles within each window. By combining the prediction results with the anomaly detection model, the time periods with future usage peaks or troughs can be identified; By comparing the predicted frequency results with the maximum service capacity of charging piles, a future usage pressure map is generated, and a frequency utilization report is output.

Images