A charging pile resource configuration method and system, and a storage medium

Abstract

The application provides a charging pile resource configuration method and system and a storage medium, relates to the technical field of charging pile resource configuration, and the method comprises the following steps: obtaining historical charging data and historical date labels of a target service area; dividing historical dates into non-holiday periods and holiday periods, constructing a holiday stage identification model based on the date distribution in the holiday period, and the model is used for dividing the holiday period into at least two holiday sub-stages; identifying the holiday sub-stage to which a target date belongs based on the target date and the holiday stage identification model; calling a prediction model corresponding to the holiday sub-stage to which the target date belongs to predict the charging demand of the target service area in a future time window according to the holiday sub-stage to which the target date belongs, and outputting a prediction result; generating a resource scheduling scheme according to the prediction result, and outputting a resource configuration instruction according to the resource scheduling scheme. The application can improve the pertinence and accuracy of charging demand prediction.

Description

Technical Field

[0001] This application relates to the technical field of charging pile resource allocation, and in particular to a charging pile resource allocation method, system and storage medium. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the construction scale of charging infrastructure in highway service areas continues to expand. However, charging demand exhibits a significant tidal characteristic in terms of time: on weekdays, the utilization rate of charging piles in service areas is generally low, resulting in poor operational efficiency; while during peak travel periods such as holidays, service areas frequently experience queuing and congestion due to a shortage of charging piles and an abundance of vehicles.

[0003] Chinese invention patent application publication number CN118798574A, published on October 18, 2024, discloses a method for planning the number of charging piles. This patent constructs a demand database by acquiring vehicle operation status data over historical periods, and determines multiple influencing factors such as charging demand distribution vector, holiday factor vector, and seasonal factor vector, and uses a time series prediction algorithm to calculate the planned number of charging piles.

[0004] However, the aforementioned patents treat holidays as a whole or a single scenario, using uniform prediction parameters or model structures to predict charging demand regardless of how many days the holiday lasts. This fails to identify the differences in charging demand at different stages within the holiday, resulting in insufficient prediction accuracy and consequently affecting the rationality and effectiveness of resource allocation. Summary of the Invention

[0005] To improve the targeting and accuracy of charging demand forecasting, this application provides a charging pile resource allocation method, system, and storage medium.

[0006] Firstly, this application provides a method for allocating charging pile resources, which adopts the following technical solution: A method for allocating charging pile resources includes the following steps: Obtain historical charging data and corresponding historical date tags for the target service area; Based on historical charging data and corresponding historical date tags, historical dates are divided into non-holiday periods and holiday periods. A holiday phase identification model is constructed based on the date distribution within the holiday period. The holiday phase identification model is used to divide the holiday period into at least two holiday sub-phases, which include the travel period and the return period. Based on the target date and holiday phase identification model, identify the holiday sub-phase to which the target date belongs; Based on the holiday sub-phase to which the target date belongs, the prediction model corresponding to that holiday sub-phase is invoked to predict the charging demand of the target service area within the future time window, and the prediction results are output. Generate a resource scheduling plan based on the prediction results, and output resource configuration instructions based on the resource scheduling plan.

[0007] This application obtains historical charging records with date tags from the target service area. First, it preliminarily categorizes dates into non-holiday and holiday periods, removing interference data from daily commuting charging. Then, based on the date distribution within holiday periods, it constructs a stage identification model that further divides holidays into different sub-stages. This application changes the traditional, crude approach of using a single model to predict all dates across all dates, accurately distinguishing differences in user charging behavior across different holiday stages and improving the targeting and accuracy of charging demand prediction. By identifying specific sub-stages and calling dedicated prediction models, this application can precisely pinpoint the steepness of the charging demand curve and the timing of extreme values ​​within different sub-stages. Through resource allocation commands, it can ensure that a specified number of charging stations are available in the service area at a specified time.

[0008] In this way, each service area can dynamically adjust the number of charging piles within the service area according to resource allocation instructions. If the predicted demand for charging piles in a certain service area shows an upward trend, charging piles from other service areas with a predicted downward trend can be allocated to that service area. This avoids the equipment idle loss and power waste caused by premature deployment of charging piles, while also avoiding the queuing and congestion problems caused by delayed deployment of charging piles. This improves the utilization rate of mobile charging piles and saves energy.

[0009] Optionally, the process of constructing the holiday phase identification model includes: Obtain hourly historical charging data for multiple historical holiday periods and construct a time series of charging data for each historical date; Cluster analysis was performed on the charging data time series to identify the time distribution pattern of charging data during each historical holiday period and the temporal characteristics of the charging data time series data in each cluster. Based on the time distribution pattern, the time boundary points of each holiday sub-stage are determined, and the corresponding holiday sub-stage labels are labeled for each time-series feature based on the time boundary points, which are used as the output labels of the training samples. The time-series features are used as input features for training samples to train the classification model, thus obtaining the holiday phase recognition model.

[0010] In constructing the holiday phase identification model, this application collected hourly charging frequency and power data during historical holidays. Based on this data, a fine-grained daily charging behavior time series was constructed, and a clustering algorithm was introduced to perform unsupervised grouping of the series. Through cluster analysis, this application can automatically uncover implicit temporal distribution patterns such as the number of peaks, peak occurrence times, and duration of nighttime lows for each historical date within a 24-hour period, without the need for manually defining the date boundaries for travel or return trips.

[0011] Subsequently, this application uses temporal features as input and clustered sub-stage labels as output to train a supervised classification model. This model learns the similarity and difference between charging behavior waveforms, enabling subsequent stage attribution judgments for any new target date to be based on the degree of conformity between the charging behavior waveform and typical waveforms of historical travel and return periods. This improves the stage division's resistance to interference and adaptability to changes in real user behavior. In this way, the probability of boundary misjudgment can be reduced. During off-peak hours at night, unnecessary charging pile relocation can be minimized; and when encountering true date boundaries, charging pile relocation and allocation can be more accurate, thereby improving charging pile utilization.

[0012] Optionally, after constructing the charging data time series for each historical date, the method further includes: Set the number of holiday sub-stages to be divided as K, and determine the number of time boundary points required as K-1; Based on the multi-threshold big law method, all possible K-1 combinations of boundary points in the charging data time series are traversed, and the inter-class variance of the K holiday sub-stages under each combination is calculated respectively. The optimal time boundary point is determined by combining the boundary points corresponding to the maximum value of the inter-class variance. Based on the optimal time boundary point, the historical holiday period is divided into K holiday sub-stages.

[0013] When dividing historical holidays, this application introduces the Otsu multi-threshold extension to replace empirical manual truncation. This application iterates through all possible combinations of time boundaries, dividing the total holiday duration into K sub-intervals. It then calculates the density of charging data within each interval (small intra-class variance) and the dispersion of charging data between intervals (large inter-class variance) under each combination. Maximizing the inter-class variance is used as the sole objective function, enabling the quantification and comparison of hundreds of boundary combinations. The optimal time boundary is ultimately determined to fall precisely at the inflection point where charging demand experiences a substantial surge or drop, rather than at a stable transition point. This ensures that the boundary between the travel period and the return period has the highest statistical discriminative value, improving the purity of the training sample labels.

[0014] Since the allocation of charging stations takes time, this arrangement can stagger the allocation time with the travel time, further improving the utilization rate of charging stations; moreover, allocating charging stations at turning points can reduce the chaos of mobile charging stations in the service area and improve allocation efficiency.

[0015] Optionally, a dynamic programming algorithm is used to determine the optimal time boundary point, specifically as follows: For different holiday periods, construct decomposition arrays respectively. The element in the m-th row and n-th column of the decomposition array represents the maximum inter-class variance when n time boundary points are selected in the first m days. Set the value of the element in the decomposition array to 0 when n=0. Let m be the total duration of the holiday period, and n be the combination of the dividing points corresponding to the maximum value of the K-1 elements. Then, backtrack to obtain K-1 optimal dividing points.

[0016] This application addresses the combinatorial explosion problem that may be encountered when using the big law to solve multiple thresholds. It adopts a dynamic programming algorithm for acceleration and optimization, which can theoretically ensure that the global optimal solution of maximizing inter-class variance is not overlooked, and can also significantly reduce the originally exponentially increasing computational complexity to the polynomial level. This enables the rapid segmentation of long-term holidays in charging pile edge computing or cloud scheduling systems.

[0017] Optionally, after generating the resource scheduling scheme, the method further includes: Acquire real-time traffic flow data, dynamically adjust the resource scheduling scheme based on the real-time traffic flow data, and generate a revised resource scheduling scheme. The dynamic correction includes: determining a correction coefficient based on the real-time traffic flow data, adjusting the opening and closing time window of the charging piles in the resource scheduling scheme with the correction coefficient, wherein the absolute value of the correction coefficient is not greater than a preset correction dead zone threshold, and the single correction magnitude does not exceed a preset maximum correction step size.

[0018] After generating the initial resource scheduling plan, the number of vehicles requiring charging within the service area will still fluctuate. If there are few vehicles, keeping the charging stations open will increase energy consumption. This application further uses real-time traffic flow data to dynamically correct deployment instructions and introduces a correction dead zone threshold to filter out noise interference caused by random traffic flow fluctuations. When the absolute value of the correction coefficient caused by traffic flow fluctuations is not greater than this threshold, the correction is rejected to prevent frequent plan jitter. This application also limits the magnitude of a single adjustment of the opening and closing time window by setting a maximum correction step size, improving the smoothness of the correction instructions. Thus, when encountering small fluctuations in the number of charging vehicles, the charging stations can remain open or closed for a longer period, minimizing the risk of sudden large-scale start-ups and shutdowns due to excessive single corrections, and extending the mechanical life of the backend switching equipment.

[0019] Optionally, resource configuration instructions are output according to the resource scheduling scheme, including: Based on the revised resource scheduling scheme, under the preset ramp rate constraint and start interval constraint, the number of charging piles required to be configured in the target service area within the future time window and the start and stop time window are determined, and resource configuration instructions are output.

[0020] When finally outputting specific executable resource configuration instructions, this application incorporates ramp rate constraints and start-up interval constraints as hard boundary conditions. The ramp rate constraint limits the total power increment of newly added charging modules allowed per unit time, while the start-up interval constraint specifies the minimum time difference between closing two charging devices or two sets of switching switches. This can transform the abstract instructions on the number of charging piles and time windows into an execution sequence that conforms to the transformer overload protection curve and safety regulations, effectively avoiding the superposition effect of inrush current caused by the concentrated closing of a large number of charging piles at the same time, and reducing the risk of a sudden drop in the voltage of the distribution network bus.

[0021] Optionally, after outputting the resource configuration instruction, the method further includes: Based on the resource allocation instructions, a resource allocation task is formulated, and based on the resource allocation task, anomaly detection of power grid status is performed to obtain the anomaly detection result; When the abnormal detection results meet expectations, the charging piles are flexibly switched on and off based on the charging station load and the abnormal detection results, under the preset ramp rate constraints and start-up interval constraints.

[0022] After outputting the final instruction, this application breaks down the instruction into resource allocation tasks and uses the task information to proactively detect anomalies in status parameters such as voltage dips and transformer load rates. When the detection results determine that the grid operating status meets the preset safety expectations, and combined with the actual active load of the current charging station, flexible switching is performed under the constraints of ramp rate and start-up interval. This effectively prevents the forced commissioning of new equipment when the grid is already under heavy load or has excessive harmonics, thus improving the actual success rate of flexible switching operations.

[0023] Optionally, after performing the flexible switching of charging piles, the method further includes: Collect actual charging load data of the target service area, and calculate the load prediction deviation rate for each time period based on the actual charging load data and the expected total load for the corresponding time period in the prediction results; Based on the load forecast deviation rate, the charging load forecast value for the target service area during the remaining period within the future time window is corrected.

[0024] This application collects actual load data and uses the prediction deviation of the already occurred period as a feedback signal to correct the prediction value of the remaining period in real time, thereby blocking the cross-period propagation path of the error and improving the prediction accuracy of the remaining period.

[0025] Secondly, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the method described in the first aspect.

[0026] Thirdly, this application provides a charging pile resource allocation system, which adopts the following technical solution: A charging pile resource allocation system includes: a processor and a memory communicatively connected to the processor; The memory is provided with the computer-readable storage medium described in the second aspect, and the computer-readable storage medium stores a computer program. When the processor processes a computer program stored on the computer-readable storage medium, it implements the method described in the first aspect.

[0027] In summary, this application includes at least one of the following beneficial technical effects: 1. This application obtains historical charging records with date tags in the target service area, firstly filters the dates into non-holiday and holiday categories to remove interference data from daily commuting charging, and can construct a stage recognition model that can further divide holidays into different holiday sub-stages based on the date distribution during holiday periods.

[0028] 2. This application changes the traditional, crude approach of making one-size-fits-all predictions for all dates using a single model. It can accurately distinguish the differences in user charging behavior during different holiday periods and improve the targeting and accuracy of charging demand prediction.

[0029] 3. By identifying the specific sub-stage and calling a dedicated prediction model, this application can accurately pinpoint the steepness of the charging demand curve and the timing of extreme values ​​within different sub-stages. This allows resource allocation instructions to minimize equipment idleness and power waste caused by early deployment during the travel period, as well as queuing and congestion problems caused by delayed deployment during the return period. Attached Figure Description

[0030] Figure 1 This is a flowchart of Embodiment 1 of this application; Figure 2 This is a flowchart of S22 clustering in Embodiment 1 of this application; Figure 3 This is a flowchart of the S5 resource configuration in Embodiment 1 of this application; Figure 4 This is a flowchart of the S6 switching process in Embodiment 1 of this application. Detailed Implementation

[0031] The following combination Figures 1 to 4 This application will be described in further detail.

[0032] The demand for charging stations varies across different service areas during different periods of the holiday. For example, at the beginning of the holiday, traffic flow from densely populated cities to tourist cities increases, leading to a higher demand for charging stations at service areas in those directions. Conversely, at the end of the holiday, traffic flow from tourist cities to densely populated cities increases, again resulting in a higher demand for charging stations at service areas in those directions. By dynamically adjusting the number of charging stations in different service areas at different times of the holiday, while keeping the total number of charging stations constant, can effectively improve the utilization rate of charging stations and reduce the probability of long queues for charging new energy vehicles.

[0033] When adjusting the number of charging stations in a service area, there are certain requirements regarding the timing and sensitivity of the adjustment. If a service area deploys a large number of charging stations too early, before the demand for charging stations increases, it will lead to idle equipment losses and wasted power resources. If the service area fails to replenish the deployment of charging stations in a timely manner, it will lead to queuing and congestion problems. If the adjustment is too sensitive, it will cause charging stations to be frequently moved between different service areas, which will reduce the utilization efficiency of the charging stations and shorten their lifespan. If the adjustment is not sensitive enough, it will lead to increased charging pressure in some service areas and the phenomenon of queuing for charging.

[0034] After the charging stations are scheduled, the number of vehicles needing to charge and the interval between charging stops will still fluctuate. This means that there will be times when the supply of charging stations exceeds the demand. If all charging stations remain operational during this period, it will shorten their lifespan and increase energy consumption. If idle charging stations are shut down immediately when there is an oversupply, they will need to be restarted when the number of vehicles needing to charge increases in a localized area. Frequent opening and closing of charging stations will shorten their lifespan and also impact the transformers in the service area.

[0035] Once the charging piles have completed the scheduling between service areas, they cannot be connected to the power grid at the same time, as this would impact the power grid. Therefore, it is necessary to connect the charging piles to the power grid gradually according to the operating status of the power grid, in order to avoid the superposition effect of inrush current caused by the simultaneous closing of a large number of charging piles at the same time, and reduce the risk of a sudden drop in the voltage of the distribution network bus.

[0036] To solve the above-mentioned technical problems, the specific technical solution of this application is as follows.

[0037] Example 1: This example discloses a method for allocating charging pile resources, referring to... Figure 1The method includes: S1 data acquisition, S2 modeling, S3 identification, S4 prediction, S5 resource allocation, and S6 deployment. First, historical charging data and corresponding historical date tags for the target service area are acquired. Then, based on the historical charging data and corresponding historical date tags, the historical dates are divided into non-holiday periods and holiday periods. A holiday phase identification model is constructed based on the date distribution within the holiday periods, which is used to divide the holiday periods into at least two holiday sub-phases. Next, based on the target date and the holiday phase identification model, the holiday sub-phase to which the target date belongs is identified. Then, based on the holiday sub-phase to which the target date belongs, the prediction model corresponding to that holiday sub-phase is invoked to predict the charging demand of the target service area within a future time window, and the prediction result is output. Finally, a resource scheduling scheme is generated based on the prediction result, and a resource allocation instruction is output based on the resource scheduling scheme. The execution process of each step in this embodiment is as follows: S1 data acquisition obtains historical charging data for the target service area from the charging pile operation management platform or the power grid dispatch system. The historical charging data includes: historical charging power, charging amount, charging start time, charging end time, and number of vehicles charging for each charging pile.

[0038] Obtain the historical date tags corresponding to each historical date in the historical charging data. The historical date tags include whether the date is a holiday, the specific name of the holiday (such as Spring Festival, National Day, Qingming Festival, etc.), and the position of the date in the holiday (such as the first day of the holiday, the second day of the holiday, etc.).

[0039] S2 modeling divides historical dates into non-holiday periods and holiday periods based on the historical date tags. The non-holiday periods include dates other than statutory holidays and their consecutive holidays, while the holiday periods include statutory holidays and their consecutive holidays.

[0040] A holiday phase identification model is constructed based on the date distribution within the holiday period. The holiday phase identification model is used to automatically divide the holiday period into at least two holiday sub-phases. In this embodiment, the holiday sub-phases include the travel period, the stable period, and the return period.

[0041] In this embodiment, the construction process of the holiday phase identification model specifically includes the following sub-steps: S21 constructs a time series by obtaining hourly historical charging data for multiple historical holiday periods and constructing a time series of charging data for each historical date.

[0042] Historical charging data is extracted from the S1 data collection system, covering various service areas during multiple holiday periods. Using hours as the time granularity, the total charging volume for each hour is calculated to form a charging data time series for each historical date. For example, for a specific day during a holiday, a charging data time series containing 24 data points (corresponding to 0:00 to 23:00) can be obtained. ,in, Indicates the first Hours of charging power.

[0043] In this embodiment, a holiday period with a total of no less than five days, such as the Spring Festival, Labor Day, and National Day, is selected from 2023 to 2025. A 24-dimensional charging data time series is constructed for each day within each holiday period.

[0044] In other embodiments, for holiday periods with a total number of days less than five, such as Qingming Festival and Dragon Boat Festival, since the holiday period is short, the hourly charging data of all dates within the holiday period can be spliced ​​together in chronological order to construct a one-dimensional charging data time series of length 24×D (D is the number of holiday days) to preserve the continuous changing trend of charging demand during the holiday.

[0045] S22 clustering, refer to Figure 2 Cluster analysis was performed on the charging data time series to identify the time distribution pattern of charging data during each historical holiday period and the temporal characteristics of the charging data time series data in each cluster.

[0046] Using the daily charging data time series of all historical holiday periods as samples, clustering algorithms (such as K-means algorithm, hierarchical clustering algorithm or DBSCAN algorithm) are used for cluster analysis. The clustering algorithm groups dates with similar daily charging load variation patterns into the same cluster based on the shape similarity of the charging data time series curves.

[0047] This embodiment takes a holiday period with a total of no less than five days (such as National Day) as an example, and uses the K-means clustering algorithm with a preset number of clusters K=3, corresponding to the three categories of travel period, stable period and return period. The specific process is as follows: S221 initialization: three charging data time series are randomly selected as the initial cluster centers.

[0048] Cluster center 1: A 24-dimensional vector of a randomly selected day (e.g., the first day of a holiday); Cluster center 2: A 24-dimensional vector of another randomly selected day (e.g., the 4th day of a holiday); Cluster center 3: A 24-dimensional vector of a randomly selected day (e.g., the 7th day of a holiday).

[0049] S222 allocation assigns each day to the nearest cluster center.

[0050] Calculate the Euclidean distance from each 24-dimensional vector to the three cluster centers, and assign it to the nearest cluster center; For example, in this embodiment, the distance from the curve on the second day of the holiday to cluster center 1 is 15, the distance to cluster center 2 is 80, and the distance to cluster center 3 is 92. Therefore, the curve on the second day of the holiday is assigned to cluster center 1.

[0051] S223 update: recalculate the center of each cluster.

[0052] For each cluster, the average of all date vectors within the cluster is taken point by point over time and used as the new cluster center. For example, the new cluster center 1 is the average power of all dates within the cluster at 0, 1, ..., 23. S224 iteration, repeating S222 allocation and S223 update, until the cluster center no longer changes or the change is less than a preset change value (e.g., 10). -4 After that, three cluster centers are obtained, and each cluster center is the time distribution pattern of that cluster (i.e., the typical daily load curve).

[0053] In this embodiment, the time series features of the charging data time series data of each cluster are extracted from the time distribution pattern corresponding to each cluster. The time series features include, but are not limited to, at least one of the following: total daily charging amount (the sum of the charging amount in 24 hours on a given day), peak time (the time interval in which the maximum charging power occurs on a given day, such as 8:00-9:00), peak coefficient (the ratio of daily peak power to daily average power), daily load rate (the ratio of daily average power to daily peak power), and charging demand time variability rate (the average amplitude of the change in charging power between adjacent time periods).

[0054] S23 determines the time boundary points. Based on the time distribution pattern obtained from the cluster analysis in S22, and combined with the arrangement order of each cluster on the holiday time axis, the time boundary points of each holiday sub-stage are determined.

[0055] In this embodiment, the Otsu multi-threshold method is used to determine the time boundary point, specifically including: The number of holiday sub-stages to be divided is set to K (K=3 in this embodiment), and the number of time boundary points required is determined to be K-1=2.

[0056] Based on the multi-threshold big law method, all possible K-1 combinations of boundary points in the charging data time series are traversed, and the inter-class variance of the K holiday sub-stages under each combination is calculated. Specifically, for a holiday with a total duration (total days or total hours) of N, all holidays satisfying the big law method are traversed. Combination of time boundaries ,in, This is the location of the first time boundary point. This is the location of the second time boundary point. For the first The location of each time boundary point. For each combination, the holiday time series is divided into K consecutive segments, each segment corresponding to a holiday sub-stage, and the inter-class variance among the K segments is... The calculation model is shown below:

[0057] in, For the first The weight of each segment (i.e., the proportion of the segment's duration to the total duration). For the first The average daily charging demand of each segment; This represents the global average of the total daily charging demand during the entire holiday period.

[0058] The optimal time boundary point is determined by combining the boundary points corresponding to the maximum value of the inter-class variance.

[0059] Based on the optimal time boundary point, the historical holiday period is divided into K holiday sub-stages.

[0060] In this embodiment, to reduce the computational complexity of traversal calculations, a dynamic programming algorithm is used to determine the optimal time boundary.

[0061] For different holiday periods, construct a dynamic programming array, denoted as the decomposition array, where the element in the m-th row and n-th column of the decomposition array is... This represents the maximum inter-class variance when selecting n time boundaries within the previous m days, and sets... That is, when no boundary point is selected, the inter-class variance is 0.

[0062] The time boundary point is inferred using the following recursive calculation model:

[0063] in, Indicates the first Heavenly Supreme The inter-class variance contribution value when a day is treated as an independent stage. An independent stage refers to a continuous range of days in the current subproblem of dynamic programming that is regarded as an independent holiday sub-stage, and the inter-class variance contribution is calculated when the stage is treated as a class on its own. Indicates preceding Selected from the sky The maximum inter-class variance that can be achieved at the boundary point; This is for calculating the maximum value.

[0064] Pick The value is N, representing the total duration of the holiday period. The value is elements The combination of the dividing points corresponding to the maximum value is obtained by backtracking sequentially. The optimal dividing point.

[0065] The S24 model is trained by labeling each time-series feature with a corresponding holiday sub-stage label based on the time boundary point, using the time-series feature as the input feature of the training sample, and training the classification model to obtain the holiday stage recognition model.

[0066] Based on the time boundary points determined in S23, each date within the historical holiday period is labeled with a corresponding holiday sub-stage label, which serves as the output label for the training samples. The temporal features of each date extracted from the clustering in S22 are used as the input features for the training samples. The input features and their corresponding output labels are combined into a single training sample, and all training samples are integrated into a training sample set.

[0067] A classification model constructed using a classification algorithm (such as decision tree, random forest, support vector machine, XGBoost, or neural network) is trained using a training sample set to obtain a trained classification model. This trained classification model is then used as the holiday phase identification model. This model can, for any input date (i.e., target date), output the holiday sub-phase category to which the date belongs based on the date's temporal characteristics. In this embodiment, the XGBoost algorithm is used as the classification model, with a maximum depth of 6, a learning rate of 0.1, and 100 iterations. The training sample set is divided into a training set and a validation set in a 7:3 ratio, and five-fold cross-validation is used to evaluate the model's performance.

[0068] S3 identification, based on the target date and holiday phase identification model, identifies the holiday sub-phase to which the target date belongs.

[0069] When charging demand forecasting is required, the expected time-series characteristics of the target date are obtained based on the holiday period to which the target date falls and the corresponding historical holiday periods. Specifically: The holiday period to which the target date belongs is marked as the target holiday period. The target holiday period is then mapped to historical holiday periods. The calculation model for the mapping is as follows:

[0070] in, Mapping the target date to the historical holiday period; The position of the target date within its respective holiday period; This represents the total number of days during historical holiday periods. The total number of days during the target holiday period; This is for rounding operations.

[0071] Since the same holiday (such as National Day) may last for different numbers of days in different years, it is meaningless to align directly by calendar date. By adopting the above calculation model, this embodiment can map the target date to its relative position in the same historical holiday period in order to obtain the typical time series characteristics at that position.

[0072] For example, if the target holiday duration is 7 days and the historical holiday duration is 9 days, the calculation for the 4th day of the target holiday using the above calculation model is as follows:

[0073] This means that the target holiday day 4 is mapped to the historical holiday day 5.

[0074] The historical time series features corresponding to the mapped date are used as the expected time series features of the target date. The expected time series features are then input into the holiday phase identification model to obtain the holiday sub-phase to which the target date belongs.

[0075] S4 prediction, based on the holiday sub-period to which the target date belongs, calls the prediction model corresponding to that holiday sub-period to predict the charging demand of the target service area in the future time window and outputs the prediction results.

[0076] Different prediction models are pre-trained for different holiday sub-stages, and each prediction model uses different prediction parameters or model structures to adapt to the charging demand characteristics of different stages.

[0077] When training the prediction model, the feature vector constructed based on the time-series characteristics (including daily total charging amount, peak time, peak coefficient, daily load rate, and charging demand time-varying rate) in the historical dates is used as input, and the actual charging amount corresponding to the historical date is used as the output label to train the prediction model.

[0078] In this embodiment, the prediction model corresponding to the travel period uses a Long Short-Term Memory (LSTM) network. The temporal features of the target date are input into the model, and the model outputs the charging amount of the target service area within a future time window (e.g., the next 24 hours). In this embodiment, the input layer of the LSTM network has a dimension of 5, the number of LSTM layers is 2, each LSTM layer includes 128 neurons, the activation function is tanh, and the output layer has a dimension of 1, with the activation function being Linear.

[0079] The prediction model for the return period adopts a gated recurrent unit network (GRU). In this embodiment, the input layer dimension of the gated recurrent unit network (GRU) is 5, the number of GRU layers is 2, each GRU layer includes 64 neurons, the activation function is tanh, and the output layer dimension is 1.

[0080] The prediction model for the stable period adopts an interval estimation model based on the statistical characteristics of the same historical period, specifically: Obtain daily charging data during stable periods in the past M similar holidays, calculate the sample mean and sample standard deviation. In this embodiment, the sample mean is used as the baseline prediction value, and the prediction interval is constructed using the sample mean ± 2 times the sample standard deviation. If the upper bound of the prediction interval is lower than the preset resource trigger threshold, it is determined that no new charging piles are needed. If the lower bound of the prediction interval is higher than the preset resource trigger threshold, a resource scheduling plan is generated according to the excess ratio. The calculation model is as follows:

[0081] Otherwise, maintain the current resource allocation.

[0082] For example, if a service area has 10 DC fast charging piles, and the maximum daily service capacity of a single pile is 300 kWh, then the resource trigger threshold is 3000 kWh. Exceeding 3000 kWh means operating at full capacity and additional resources are needed.

[0083] We selected the past three National Day holidays (M=3), with a stable period of 2 days in each holiday, resulting in a total of 6 sample days of charging volume: ; The sample data is shown below: [2150,2280,2320,2090,2210,2430]; The sample mean is approximately 2246.7 kWh, and the sample standard deviation is approximately 122.7 kWh.

[0084] In this embodiment, a 95% confidence level is used, and the corresponding z-value is: z 0.025 =1.96, then the prediction range is [2006.2, 2487.2] kWh, meaning that there is a 95% probability that the daily charging amount during the current stable period will fall within the prediction range.

[0085] In this embodiment, the upper limit of the prediction range of 2487.2 kWh is lower than the resource trigger threshold of 3000 kWh. Even if the daily charging volume during the stable period is at its maximum value within the normal fluctuation range, the existing charging piles can fully handle it, so there is no need to add new charging piles.

[0086] After identifying the holiday sub-period to which the target date belongs, the system automatically calls the prediction model corresponding to that holiday sub-period, inputs the daily charging data of the holiday sub-period to the prediction model in sequence, and outputs the prediction results, which include the predicted charging amount for each future time period.

[0087] S5 resource configuration generates a resource scheduling plan based on the prediction results, and outputs resource configuration instructions based on the resource scheduling plan.

[0088] Based on the prediction results and the existing number of charging piles in the target service area, a resource scheduling plan is generated. This plan includes information such as the number of charging piles required to be configured within a future time window and the activation / deactivation time window for each charging pile. The number of charging piles is used to inform maintenance personnel to deploy charging piles in advance (e.g., to schedule mobile charging piles in advance) or to automatically trigger activation / deactivation commands for each charging pile. The activation / deactivation time window for each charging pile is used to control the activation and deactivation of charging piles.

[0089] Reference Figure 3 In this embodiment, after generating the resource scheduling scheme, the method further includes: S51 Correction: Obtain real-time traffic flow data, dynamically correct the resource scheduling scheme based on the real-time traffic flow data, and generate a corrected resource scheduling scheme.

[0090] This embodiment collects real-time image data of the charging area by deploying camera devices at the entrance of the charging area and charging pile positions. A pre-trained vehicle detection model (such as YOLOv8 or Faster R-CNN) is used to detect vehicle targets in the image frames uploaded by the camera devices, identify vehicle targets in the images, and output their bounding box positions.

[0091] The system tracks and matches the same vehicle target between adjacent frames to determine the vehicle's motion state and parking space occupancy status, thus distinguishing between waiting vehicles and vehicles currently charging. In this embodiment, the rules for determining waiting and charging vehicles are as follows: vehicles located at charging piles with their charging guns connected are marked as charging vehicles; vehicles located within the charging area but not at charging piles and in a stationary state are marked as waiting vehicles. The number of waiting vehicles and the number of charging vehicles are counted separately to obtain real-time vehicle statistics for the charging area.

[0092] The sum of the number of waiting vehicles and the number of vehicles charging is used as the real-time vehicle density index of the charging area. Historical vehicle density indices for the charging area during the same period in the past (e.g., the same time during the past three similar holidays) are obtained. The vehicle density deviation rate is calculated based on the real-time vehicle density index and the historical vehicle density index. The calculation model is as follows:

[0093] in, Vehicle density deviation rate; This is a real-time vehicle density indicator. This represents the average historical vehicle density index for the charging area during the same period in history.

[0094] According to the vehicle density deviation rate The correction coefficients are determined using the following calculation model. .

[0095]

[0096] in, This is a correction factor; Vehicle density deviation rate; The preset dead zone threshold is set (e.g., 0.15, meaning no correction is performed when the vehicle density deviation rate is within ±15%). The preset scaling factor (e.g., 0.5); Calculated for absolute values.

[0097] With the aforementioned correction coefficient In this embodiment, the adjustment method for the start-up and shutdown time windows of charging piles in the resource scheduling scheme is as follows: if the correction coefficient... A value greater than 0 indicates that the current vehicle density in the charging area is higher than the historical average for the same period, and the opening and closing time window needs to be advanced. minutes; if the correction factor A value less than 0 indicates that the current vehicle density in the charging area is lower than the historical average for the same period, and the opening and closing time window needs to be postponed. minutes. Where T is the preset base time window length (e.g., 60 minutes).

[0098] To ensure the stability of the adjustment, the magnitude of a single correction is limited to no more than the preset maximum correction step size. ,Right now:

[0099] in, This is the final correction amount, in minutes; For sign function, correction coefficient Greater than 0, The value is +1, if the correction factor is... If less than 0, then The value is -1; This is an operation to find the minimum value; Correction coefficient The absolute value; The length of the base time window; The maximum correction step size is set to 30 minutes in this embodiment, but can be set according to requirements in other embodiments.

[0100] For example, if the number of waiting vehicles in the current charging area is 8 and the number of vehicles charging is 12, totaling 20 vehicles, the average vehicle density at the same time in history is 15 vehicles.

[0101] Vehicle density deviation rate If the value exceeds the preset dead zone threshold by 15%, and a proportionality coefficient A = 0.5 is used, the correction coefficient is calculated. If the base time window is 60 minutes, the correction amount is about 10 minutes. If it does not exceed the maximum correction step of 30 minutes, the opening and closing time window will be advanced by 10 minutes.

[0102] Through the above dynamic correction, this embodiment uses the real-time perceived vehicle density information in the charging area as the input for resource allocation adjustment, enabling the resource scheduling scheme to quickly respond to real-time changes in actual charging demand.

[0103] S52 outputs instructions to determine the number of charging piles required to be configured in the target service area within the future time window and the start-up and closing time window of each charging pile, based on the modified resource scheduling scheme and under preset ramp rate constraints and start-up interval constraints, and outputs resource configuration instructions.

[0104] The resource configuration instructions include the target service area identifier, the number of charging piles to be deployed in advance during the holiday period, the start-up and shutdown time windows of each charging pile during the holiday period, the power configuration parameters of the charging piles, and the start-up time interval of each charging pile. The ramp rate constraint refers to the maximum rate of change of a single charging pile from zero power to the target power (e.g., not exceeding 5kW / s), and the start-up interval constraint refers to the minimum time interval between two adjacent charging piles when multiple charging piles are started (e.g., not less than 15 seconds).

[0105] This embodiment, by setting ramp rate constraints and start-up interval constraints, can minimize the severe impact on the power grid caused by the concentrated start-up of multiple charging piles.

[0106] S6 switching enables flexible switching of charging piles based on resource allocation instructions.

[0107] Reference Figure 4 After outputting the resource configuration command, the method further includes the following sub-steps: S61 Anomaly Detection: Based on the resource configuration instruction, a resource configuration task is formulated, and based on the resource configuration task, anomaly detection of power grid status is performed to obtain the anomaly detection result.

[0108] Based on the resource configuration instructions output in the S5 resource configuration, specific resource configuration tasks are generated. The resource configuration tasks include parameters such as the start time, target power, ramp rate, and start interval of each charging pile.

[0109] Before executing the resource allocation task, this embodiment also performs power grid status anomaly detection based on the resource allocation task. The anomaly detection includes: transformer load rate detection and voltage sag detection.

[0110] Transformer load rate detection refers to calculating the expected total load, and the calculation model for the expected total load is as follows:

[0111] in, For the projected total load; The current total load of the target service area is the real-time load data of the charging stations in the target service area, including the number and power of currently operating charging piles, and the power load of auxiliary equipment such as station lighting and air conditioning. The planned new load refers to the total active power of the charging piles to be put into operation in this project, based on the resource allocation task.

[0112] If the total load is expected to exceed the transformer's rated capacity With preset safety factor The product of (in this embodiment, its value is 0.85) indicates a risk of transformer overload; if the expected total load does not exceed the transformer's rated capacity and the preset safety factor, then... If the product of the two values ​​is found, then it is determined that there is no risk of transformer overload.

[0113] Voltage sag detection refers to calculating the expected voltage drop (e.g., 5%) after a planned increase in load is added, based on line impedance parameters. The calculation model is as follows:

[0114]

[0115] in, This represents the expected voltage drop value. The planned new load is the total active power of the charging piles to be put into operation this time, according to the resource allocation task; R is the line resistance, and is the real part of the line impedance. The new reactive power is the reactive power component when new charging piles are planned to be put into operation; X is the line reactance, which is the imaginary part of the line impedance. The power factor angle of the charging pile; For inverse cosine calculation; It is the tangent function; This is the current effective voltage value.

[0116] If none of the above detection indicators exceed the corresponding preset safety threshold, the abnormal detection result is determined to be in line with expectations, and S62 is executed in line with expectations; if any indicator exceeds the preset safety threshold, the abnormal detection result is determined to be in line with expectations.

[0117] S62 meets expectations. When the anomaly detection results meet expectations, the charging piles are flexibly switched on and off based on the charging station load and the anomaly detection results, under the preset ramp rate constraints and start-up interval constraints.

[0118] If the abnormal detection results show that the voltage sag and transformer load rate do not exceed the preset safety threshold, then the abnormal detection results are deemed to be in line with expectations, and the flexible switching of charging piles is directly executed according to the ramp rate constraint and start-up interval constraint determined in the S5 resource configuration.

[0119] When abnormal detection results do not meet expectations, an early warning signal is issued to the operation and maintenance personnel.

[0120] After the flexible switching of charging piles is performed, the actual charging load data of the target service area is collected. The actual charging load data includes the sum of the real-time power of all charging piles in operation, as well as the power load of auxiliary equipment such as station lighting and air conditioning.

[0121] Based on the actual charging load data and the predicted total load for the corresponding time period in the prediction results, the load prediction deviation rate for each time period is calculated, specifically as follows: The future time window is divided into several consecutive time periods (e.g., each time period is 15 minutes). For each completed time period, the actual charging load data within that time period is calculated and compared with the predicted total load for the corresponding time period in the forecast results. The load forecast deviation rate for that time period is calculated, and the calculation model for the load forecast deviation rate is as follows:

[0122] in, Load forecast deviation rate; This is actual charging load data; This represents the estimated total load for the corresponding time period based on the actual charging load data.

[0123] Based on the load forecast deviation rate, the resource allocation task for the target service area during the remaining time period within the future time window is adjusted, specifically as follows: The corrected model for the number of charging piles is shown below:

[0124] in, This is a correction amount for the number of charging stations; This represents the number of charging stations to be activated during the remaining time period. This represents the load forecast deviation for the current time period. The single adjustment step size for the number of charging piles is, in this embodiment, 20% of the remaining number to be activated; The function is for finding the minimum value; It is a rounding function; This is for absolute value operations.

[0125] The corrected number of charging piles is obtained, and the corrected number of charging piles is constrained by the number of pre-deployed charging piles. That is, if the number of charging piles in the corrected result exceeds the number of pre-deployed charging piles, the corrected result is taken as the number of pre-deployed charging piles; otherwise, the corrected result remains the original value.

[0126] Example 2: This example provides a computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the method described in Example 1.

[0127] Example 3: This example discloses a charging pile resource allocation system, the system including: a processor, and a memory communicatively connected to the processor; The memory is provided with a computer-readable storage medium as described in Embodiment 2, and a computer program is stored on the computer-readable storage medium. When the processor processes the computer program stored on the computer-readable storage medium, it implements the method described in Embodiment 1.

[0128] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for allocating charging pile resources, characterized in that, include: Obtain historical charging data and corresponding historical date tags for the target service area; Based on historical charging data and corresponding historical date tags, historical dates are divided into non-holiday periods and holiday periods. A holiday phase identification model is constructed based on the date distribution within the holiday period. The holiday phase identification model is used to divide the holiday period into at least two holiday sub-phases. Based on the target date and holiday phase identification model, identify the holiday sub-phase to which the target date belongs; Based on the holiday sub-phase to which the target date belongs, the prediction model corresponding to that holiday sub-phase is invoked to predict the charging demand of the target service area within the future time window, and the prediction results are output. Generate a resource scheduling plan based on the prediction results, and output resource configuration instructions based on the resource scheduling plan.

2. The charging pile resource allocation method according to claim 1, characterized in that, The construction process of the holiday phase identification model includes: Obtain hourly historical charging data for multiple historical holiday periods and construct a time series of charging data for each historical date; Cluster analysis was performed on the charging data time series to identify the time distribution pattern of charging data during each historical holiday period and the temporal characteristics of the charging data time series data in each cluster. Based on the time distribution pattern, the time boundary points of each holiday sub-stage are determined, and the corresponding holiday sub-stage labels are labeled for each time-series feature based on the time boundary points, which are used as the output labels of the training samples. The time-series features are used as input features for training samples to train the classification model, thus obtaining the holiday phase recognition model.

3. The charging pile resource allocation method according to claim 2, characterized in that, After constructing the charging data time series for each historical date, the method further includes: Set the number of holiday sub-stages to be divided as K, and determine the number of time boundary points required as K-1; Based on the multi-threshold big law method, all possible K-1 combinations of boundary points in the charging data time series are traversed, and the inter-class variance of the K holiday sub-stages under each combination is calculated respectively. The optimal time boundary point is determined by combining the boundary points corresponding to the maximum value of the inter-class variance. Based on the optimal time boundary point, the historical holiday period is divided into K holiday sub-stages.

4. The charging pile resource allocation method according to claim 3, characterized in that, The optimal time breakpoint is determined using a dynamic programming algorithm, specifically as follows: For different holiday periods, construct decomposition arrays respectively. The element in the m-th row and n-th column of the decomposition array represents the maximum inter-class variance when n time boundary points are selected in the first m days. Set the value of the element in the decomposition array to 0 when n=0. Let m be the total duration of the holiday period, and n be the combination of the dividing points corresponding to the maximum value of the K-1 elements. Then, backtrack to obtain K-1 optimal dividing points.

5. The charging pile resource allocation method according to any one of claims 1-4, characterized in that, After generating the resource scheduling scheme, the method further includes: Acquire real-time traffic flow data, dynamically adjust the resource scheduling scheme based on the real-time traffic flow data, and generate a revised resource scheduling scheme. The dynamic correction includes: determining a correction coefficient based on the real-time traffic flow data, adjusting the opening and closing time window of the charging piles in the resource scheduling scheme with the correction coefficient, wherein the absolute value of the correction coefficient is not greater than a preset correction dead zone threshold, and the single correction magnitude does not exceed a preset maximum correction step size.

6. The charging pile resource allocation method according to claim 5, characterized in that, Output resource configuration instructions based on the resource scheduling scheme, including: Based on the revised resource scheduling scheme, under the preset ramp rate constraint and start interval constraint, the number of charging piles required to be configured in the target service area within the future time window and the start and stop time window are determined, and resource configuration instructions are output.

7. The charging pile resource allocation method according to claim 6, characterized in that, After outputting the resource configuration command, the method further includes: Based on the resource allocation instructions, a resource allocation task is formulated, and based on the resource allocation task, anomaly detection of power grid status is performed to obtain the anomaly detection result; When the abnormal detection results meet expectations, the charging piles are flexibly switched on and off based on the charging station load and the abnormal detection results, under the preset ramp rate constraints and start-up interval constraints.

8. The charging pile resource allocation method according to claim 7, characterized in that, After performing the flexible switching of charging piles, the method further includes: Collect actual charging load data of the target service area, and calculate the load prediction deviation rate for each time period based on the actual charging load data and the expected total load for the corresponding time period in the prediction results; Based on the load forecast deviation rate, the resource allocation task for the target service area during the remaining time period within the future time window is adjusted.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, is used to implement the method as described in any one of claims 1-8.

10. A charging pile resource allocation system, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory is provided with a computer-readable storage medium as described in claim 9, and a computer program is stored on the computer-readable storage medium. When the processor processes a computer program stored on the computer-readable storage medium, it implements the method as described in any one of claims 1-8.

Images