An electric bicycle charging scheduling method based on load prediction

Abstract

The application relates to the technical field of intelligent power distribution, in particular to an electric bicycle charging scheduling method based on load prediction, which comprises the following steps: acquiring the load of a power distribution network station area, a riding-in frequency, the rated power of each socket, the temperature of a charging area, the transient voltage of each socket, feeder impedance and spatial coordinates; calculating a global capacity limit value based on the load of the station area, the riding-in frequency and the rated power; extracting a polarization scalar based on the temperature and the transient voltage and combining the rated power to generate an environment limit value; extracting line pressure drop and sunlight exposure features based on the feeder impedance and the spatial coordinates to perform partition clustering, generating an updated limit value according to spatial geometric correlation and splitting a power quota; under the quota constraint, inputting the polarization scalar and the feeder impedance into an evaluation model to output a scheduling characteristic value to control a relay switch gear position; collecting a steady-state voltage drop to generate a degradation characteristic value, and controlling a contactor to be disconnected when the value exceeds the limit. The application realizes power safety scheduling through step-by-step quota control.

Description

Technical Field

[0001] This invention relates to the field of intelligent power distribution technology, specifically to a method for scheduling electric bicycle charging based on load prediction. Background Technology

[0002] With the increasing popularity of electric bicycles, the large-scale construction of centralized charging stations is growing. These charging stations are typically directly connected to the distribution network's substation system, and their high-density concurrent charging often causes significant load impacts on the local power grid. To ensure the safe and stable operation of the power grid and the charging experience for users, it is necessary to implement reasonable power scheduling for each charging socket within the centralized charging stations.

[0003] Currently, charging scheduling methods for electric bicycles typically monitor the total load status of the local power distribution network. This allows for overall limiting of the total output power of charging stations during peak electricity consumption periods, preventing transformer overload. Existing charging control systems generally possess basic safety monitoring mechanisms, such as static threshold comparisons using ambient temperature or single voltage parameters. When a parameter exceeds a preset safety limit, power outages or power reduction measures are implemented. Under normal operating conditions, these methods effectively maintain the basic operation and electrical safety of charging stations.

[0004] However, in actual high-density centralized charging scenarios, the safety boundary and power transmission capacity of the charging network are actually constrained by the deep coupling of multiple physical factors. On the one hand, the available capacity of the distribution network area will exhibit significant dynamic fluctuations with regular electricity consumption behavior; on the other hand, due to the vast physical space of the charging station, the actual operating conditions faced by charging sockets at different locations within the station vary greatly. For example, the socket located at the end of the power supply feeder topology has an inherent line voltage drop. If the socket is simultaneously located in an area exposed to direct sunlight without any shade and is connected to an aging battery, the superposition of multiple stresses such as "electrical line voltage drop," "local environmental heat radiation," and "battery polarization heating" can easily trigger severe thermal runaway or contact device ablation risks at local micro-nodes. Most existing conventional dispatching strategies treat grid capacity, ambient temperature, and electrical status as independent single-dimensional conditions for isolated judgment, making it difficult to effectively assess and address the hidden coupling risks brought about by the superposition of multiple physical field characteristics.

[0005] In summary, how to achieve safe and precise power allocation in charging stations and effective protection of underlying hardware is a technical problem that urgently needs to be solved in this field.

[0006] To address this, a method for scheduling electric bicycle charging based on load prediction is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a load prediction-based electric bicycle charging scheduling method that achieves safe power scheduling through tiered quota control. The method includes: acquiring the distribution network area load, entry frequency, rated power of each charging socket, charging area temperature, transient voltage of each charging socket, feeder impedance, and spatial coordinates; calculating global capacity limits based on the area load, entry frequency, and rated power; extracting polarization scalars based on temperature and transient voltage and combining them with rated power to generate environmental limits; extracting line voltage drop and sunlight exposure characteristics based on feeder impedance and spatial coordinates for partitioned clustering, generating updated limits based on spatial geometric correlations, and splitting power quotas; under quota constraints, inputting the polarization scalars and feeder impedance into an evaluation model to output scheduling feature values ​​to control relay switching positions; collecting steady-state voltage drop to generate degradation feature values, and controlling contactor disconnection when limits are exceeded.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for scheduling electric bicycle charging based on load prediction, comprising: Obtain the distribution network's substation load, entry frequency, and rated power of each socket; obtain the charging area temperature, transient voltage of each socket, feeder impedance, and spatial coordinates; calculate the global capacity limit based on the substation load, entry frequency, and rated power; extract the polarization scalar based on temperature and transient voltage, and generate environmental limits by combining them with the rated power; Based on feeder impedance and spatial coordinates, line voltage drop and solar exposure characteristics are extracted and divided into regions: dual-characteristic region, single-characteristic region, and unconstrained region. Within each region, the socket with the largest polarization scalar is extracted as the reference socket. Based on spatial coordinates, sockets whose distance from the reference socket is less than a preset distance threshold are selected to form an adjacent set. The environmental limits of the adjacent set are updated using the ratio of distance to the distance threshold to generate updated limits. When the sum of the limits of all sockets is greater than the global capacity limit, the global capacity limit is split into power quotas corresponding to each region. Under power quota constraints, the polarization scalar and the deep level and path congestion feature values ​​extracted from the topology tree constructed based on the feeder impedance are input into the classification network to output scheduling feature values, and control data is generated to control the relay to switch gears; the steady-state voltage drop after the relay is closed is collected; the difference between the steady-state voltage drop and the preset nominal voltage drop is used to generate a degradation feature value; when the degradation feature value is greater than the preset degradation extreme value, the contactor is controlled to open.

[0009] Preferably, the process of acquiring the distribution network area load, entry frequency, and rated power of each socket; and acquiring the charging area temperature, transient voltage of each socket, feeder impedance, and spatial coordinates includes: acquiring the distribution network area load and rated power of each socket through an area edge computing gateway; collecting vehicle arrival events through a visual sensing terminal deployed at the entrance of the charging area, and accumulating and calculating the entry frequency within a preset time window; collecting the charging area temperature through an environmental sensing node array; at the initial conduction moment when the electric bicycle physically connects to each charging socket, using the high-frequency sampling circuit built into the socket to capture the voltage drop and recovery waveform within a preset period as the transient voltage of each socket; extracting spatial coordinates based on a pre-constructed three-dimensional grid topology of the charging station, and quantitatively calculating the feeder impedance based on the physical wire diameter and wiring physical length from each socket to the secondary side of the area transformer.

[0010] Preferably, the process of calculating the global capacity limit and environmental limit includes: obtaining the distribution network transformer capacity and expected fluctuating load; calculating the difference between the transformer capacity and the sum of the substation load and the expected fluctuating load to generate a dynamic power supply margin; multiplying the inrush frequency, preset time window, and rated power to predict the tidal load increment; comparing the dynamic power supply margin with the tidal load increment to determine the global capacity limit; extracting the time-domain waveform features of the transient voltage to obtain the ohmic voltage drop amplitude and relaxation time constant; multiplying the ohmic voltage drop amplitude and the relaxation time constant to generate a polarization scalar; obtaining historical temperature sequences and historical transient voltages; dividing the historical temperature sequence into multiple temperature intervals according to a preset temperature step size; and extracting the historical transient voltage in each of the aforementioned temperature ranges. The average local pressure drop within each temperature range is calculated, and the temperature center point values ​​of each temperature range are spliced ​​with the corresponding average local pressure drop values ​​in a two-dimensional space to construct an impedance temperature drift matrix. Absolute temperature data is extracted from the temperature range and used as input keys. A spline interpolation algorithm is then used to perform a continuous mapping solution on the impedance temperature drift matrix to obtain the corresponding temperature pressure drop mapping value. This mapping value is divided by the system's preset standard environmental reference pressure drop and normalized to generate a dimensionless thermodynamic impedance correction factor. The ratio of the rated power divided by the thermodynamic impedance correction factor and the polarization scalar product is calculated to obtain an initial quotient. This initial quotient is compared with the rated power, and the smaller of the two is taken as the environmental limit value for each socket.

[0011] Preferably, the process of dividing the area into dual-feature area, single-feature area, and unconstrained area includes: extracting the full-load operating current of each socket under rated power; multiplying the full-load operating current with the corresponding feeder impedance to quantify and generate the theoretical maximum line loss voltage drop of each socket, which is used as the line voltage drop feature; using spatial coordinates to perform illumination tracing and shadow projection calculations in a preset three-dimensional building model of the station to extract the cumulative exposure time of each socket under direct sunlight, which is used as the sunlight exposure feature; and combining the line voltage drop feature with the sunlight exposure feature. A two-dimensional feature space is constructed based on the exposure characteristics, and extreme values ​​for voltage drop risk and exposure duration are set in the two-dimensional feature space. The set of sockets whose line voltage drop characteristics are greater than the extreme value for voltage drop risk and whose sunlight exposure characteristics are greater than the extreme value for exposure duration are divided into dual-feature regions. The set of sockets whose only line voltage drop characteristics are greater than the extreme value for voltage drop risk or whose only sunlight exposure characteristics are greater than the extreme value for exposure duration are divided into single-feature regions. The set of sockets whose line voltage drop characteristics are not greater than the extreme value for voltage drop risk and whose sunlight exposure characteristics are not greater than the extreme value for exposure duration are divided into unconstrained regions.

[0012] Preferably, the process of generating the updated limit includes: traversing all sockets in each partition and marking the socket with the highest polarization scalar value as the reference socket; calculating the Euclidean distance between each socket and the reference socket based on the spatial coordinates; filtering out sockets whose Euclidean distance is less than the preset distance threshold to construct the adjacent set; dividing the Euclidean distance corresponding to each socket in the adjacent set by the preset distance threshold to generate a spatial thermal attenuation coefficient less than 1; and multiplying the original environmental limit of each socket in the adjacent set by the corresponding spatial thermal attenuation coefficient to calculate and generate the derated updated limit.

[0013] Preferably, the process of splitting the global capacity limit into power quotas corresponding to each zone includes: accumulating the updated limits and environmental limits of all sockets within the target charging area to generate a sum of limits for each socket; when the sum of limits for each socket is greater than the global capacity limit, for the dual-feature region, applying a first-level thermal impedance attenuation and voltage drop compensation pre-deduction to the global capacity limit, extracting the corresponding capacity to generate a first stringent quota; for the single-feature region, extracting the capacity corresponding to the current region and implementing a unilateral derating constraint to generate a second restricted quota; for the unconstrained region, subtracting the first stringent quota and the second restricted quota from the global capacity limit to obtain the remaining capacity limit, assigning it to the current region as a third regular quota.

[0014] Preferably, the evaluation model includes: a topology reconstruction layer, which maps and reconstructs all charging sockets into a virtual power supply topology tree with hierarchical parent-child relationships based on the branch connection relationships in the construction wiring topology data and the feeder impedance of each charging socket, and uses the polarization scalar as the attribute feature of the corresponding node; a line environment interaction layer, which classifies each target socket in the virtual power supply topology tree as a core node, and classifies the remaining charging sockets that share the same upstream power supply path as homogeneous environment nodes; counts the number of nodes in the homogeneous environment nodes whose polarization scalar is greater than that of the core node, calculates the ratio of the number of such nodes to the total number of homogeneous environment nodes, and generates a path congestion feature value; a collaborative decision layer, which concatenates the polarization scalar of the target socket, its depth level in the virtual power supply topology tree, and the path congestion feature value into a context feature vector, inputs it into the classification network structure, and outputs the corresponding scheduling feature value; wherein, when the depth level is greater than a preset depth level and the path congestion feature value is greater than a preset congestion extreme value, the classification network structure is forced to output the scheduling feature value ranked last.

[0015] Preferably, the process of controlling the contactor to disconnect when the degradation characteristic value is greater than the preset degradation extreme value includes: mapping the scheduling characteristic value to a preset discrete power level table in descending order of value, querying and generating the control data corresponding to each socket; outputting a drive level based on the control data, controlling the working circuit relay array built into the socket to close to the physical conduction path of the corresponding level; after the physical conduction path is closed, starting a timer and continuing for a preset stable delay period, reading the voltage difference between the input and output terminals of the working circuit relay array at the end of the preset stable delay period as the steady-state voltage drop, and simultaneously collecting the actual working current on the physical conduction path; calculating the current contact resistance by dividing the steady-state voltage drop by the actual working current, calculating the absolute difference between the current contact resistance and the preset nominal contact resistance, and using the absolute difference as the degradation characteristic value characterizing the internal resistance state of mechanical contact erosion; when the degradation characteristic value is greater than the preset degradation extreme value, generating a hardware-level physical isolation instruction, and using the hardware-level physical isolation instruction to directly drive the main circuit contactor connected in series with the working circuit relay array to perform a mechanical disconnection action.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing an evaluation model that includes topology reconstruction and along-line environment interaction layers, discrete charging sockets are reconstructed into a power supply topology tree, and objective path congestion characteristics are used to assess local risks. This enables the system to accurately identify heat accumulation and voltage drop superposition on the same power supply branch, and directly force the output of derating commands when deep nodes face high congestion, effectively preventing overload caused by multiple devices charging at high power concurrently on local lines, and improving the safety of power grid end operation.

[0017] 2. A differentiated quota allocation strategy based on two-dimensional feature space clustering is proposed. The charging area is divided into different constraint zones according to the physical superposition of line voltage drop and environmental shading. By preferentially applying thermal impedance attenuation and pre-deduction to high-risk dual-feature zones and assigning remaining quotas to unconstrained zones, the system achieves structured allocation of global capacity resources.

[0018] 3. By using the absolute difference between the steady-state voltage drop and the nominal voltage drop as a quantitative basis, the system can objectively assess the degree of erosion and degradation of mechanical contacts. When the degradation parameter exceeds the limit, the main circuit contactor is directly driven to perform physical power-off isolation. This mechanism closely links the algorithm decision with the underlying hardware action, avoiding misjudgment of action due to transient interference and reducing the electrical safety hazards of the underlying hardware. Attached Figure Description

[0019] Figure 1 A schematic diagram of a method for scheduling electric bicycle charging based on load prediction provided by the present invention; Figure 2 A schematic diagram illustrating the process of calculating global capacity limits and environmental limits provided by this invention; Figure 3 This is a schematic diagram illustrating the process of dividing a region into a dual-feature region, a single-feature region, and an unconstrained region, as provided by the present invention. Detailed Implementation

[0020] 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.

[0021] Please see Figures 1 to 3 This invention provides a method for scheduling electric bicycle charging based on load prediction, the technical solution of which is as follows: A method for scheduling electric bicycle charging based on load prediction, the specific process of which is as follows: Figure 1 As shown, it includes: Obtain the load, entry frequency, and rated power of each socket in the distribution network area; obtain the temperature of the charging area, transient voltage of each socket, feeder impedance, and spatial coordinates; calculate the global capacity limit based on the load, entry frequency, and rated power of the distribution network area; extract the polarization scalar based on the temperature and transient voltage, and generate the environmental limit by combining it with the rated power; Based on feeder impedance and spatial coordinates, line voltage drop and solar exposure characteristics are extracted and divided into regions: dual-characteristic region, single-characteristic region, and unconstrained region. Within each region, the socket with the largest polarization scalar is extracted as the reference socket. Based on spatial coordinates, sockets whose distance from the reference socket is less than a preset distance threshold are selected to form an adjacent set. The environmental limits of the adjacent set are updated using the ratio of distance to the distance threshold to generate updated limits. When the sum of the limits of all sockets is greater than the global capacity limit, the global capacity limit is split into power quotas corresponding to each region. Under power quota constraints, the polarization scalar and feeder impedance are input to the evaluation model to output scheduling characteristic values. Control data is generated based on the scheduling characteristic values, and the relay is controlled to switch positions. The steady-state voltage drop after the relay is closed is collected. The difference between the steady-state voltage drop and the preset nominal voltage drop is used to generate a degradation characteristic value. When the degradation characteristic value is greater than the preset degradation extreme value, the contactor is controlled to open.

[0022] Example 1: Furthermore, the process of acquiring the distribution network area load, entry frequency, and rated power of each socket; and acquiring the charging area temperature, transient voltage of each socket, feeder impedance, and spatial coordinates includes: acquiring the distribution network area load and rated power of each socket through the area edge computing gateway; collecting vehicle arrival events through a visual sensing terminal deployed at the entrance of the charging area, and accumulating and calculating the entry frequency within a preset time window; collecting the charging area temperature through an environmental sensing node array; at the initial conduction moment when the electric bicycle physically connects to each charging socket, using the high-frequency sampling circuit built into the socket to capture the voltage drop and recovery waveform within a preset period as the transient voltage of each socket; extracting spatial coordinates based on the pre-constructed three-dimensional grid topology of the charging station, and quantitatively calculating the feeder impedance based on the physical wire diameter and wiring physical length from each socket to the secondary side of the area transformer.

[0023] To obtain the load data of the distribution network area and the rated power of each charging socket, a communication connection is established between the distribution area edge computing gateway and the smart meter or distribution monitoring terminal on the low-voltage side of the distribution network transformer. The gateway reads the current active and reactive power output data of the transformer according to a preset polling cycle, using this as the distribution network area load data. Simultaneously, the gateway extracts the maximum allowable output power setting value of each socket under standard operating conditions by parsing the device configuration messages initially sent to each charging socket by the system, using this as the rated power of the corresponding socket.

[0024] For the acquisition process of vehicle entry frequency, a visual sensing terminal deployed at the physical entrance of the charging area is activated to continuously capture the live video stream. A virtual pixel reference line is pre-marked across the physical entrance in the video frame. Using a pre-deployed convolutional neural network target detection algorithm, moving targets in the video stream are extracted frame by frame, and target bounding boxes with electric bicycle features are output. The appearance features and motion state of each target bounding box are extracted, and the data association logic in the multi-target tracking algorithm is used to assign a unique target ID to the same electric bicycle in consecutive video frames, forming a continuous motion trajectory with time sequence attributes. The center pixel coordinates of the target bounding box are continuously tracked. When the displacement vector of the center pixel coordinates in consecutive adjacent video frames is detected to intersect with the virtual pixel reference line, a cross-boundary action is determined to be completed, a vehicle arrival event is recorded, and a corresponding timestamp is generated. During continuous operation, the system maintains a sliding time window of fixed length, counts the total number of arrival events within the time window from the current moment, and then divides the total number by the time span of the time window to quantify and output the vehicle entry frequency per unit time.

[0025] To acquire the temperature of the charging area, multiple environmental sensor nodes are activated in a grid-like distribution within the physical space of the charging station. Each sensor node continuously detects the physical temperature of the surrounding air using its internal sensing elements, converts it into an electrical signal, performs analog-to-digital conversion, and periodically reports temperature and humidity values ​​with location tags to the aggregation node, forming a temperature data stream covering the entire charging area.

[0026] For the acquisition process of transient voltages in each socket, the high-frequency sampling circuit built into the socket operates continuously in the background at a high-frequency clock rate in standby mode, and continuously pushes the sampled data into a circular data buffer of a set length for rolling over. Hardware-level edge-triggered detection logic based on a voltage drop threshold is set on the electrical circuit of the socket. When the charging plug of an electric bicycle is inserted into the socket, causing physical contact at the endpoints, the negative leading edge signal generated by the circuit voltage will initially activate the trigger. To eliminate physical jitter and transient arc interference generated at the moment of mechanical contact, the system introduces a hardware and software combined de-jitter confirmation mechanism: requiring the negative leading edge signal... Not only must the voltage drop exceed a preset threshold (e.g., greater than 5% of the rated open-circuit voltage), but the state after the drop must also persist for more than a preset minimum conduction confirmation time (e.g., 20 milliseconds) on the timeline to be considered a valid electrical physical conduction event. After confirming a valid conduction event, the high-frequency sampling circuit freezes the current circular data buffer, extracts historical resident data for a preset duration before the trigger point and continuous tracking data for a preset duration after the trigger point, concatenates and reassembles the two data segments, and uses a low-pass digital filtering algorithm to smooth the reassembled data sequence to completely filter out residual high-frequency mechanical contact noise. Finally, the complete voltage drop and recovery waveform, including the initial rapid voltage drop boundary and the subsequent stabilization state, is extracted from the smoothed data and saved as the current transient voltage of the socket. After the transient voltage data is saved, the system issues a hardware reset command to clear the trigger's edge detection lock flag and unfreeze the circular data buffer, controlling the high-frequency sampling circuit to resume the background scrolling overwrite mode to await the next charging plug connection trigger event.

[0027] To address the process of obtaining spatial coordinates and feeder impedance, a two-dimensional Cartesian coordinate reference system is first established based on the actual physical plane of the charging station. The actual installation column or backplate position of each charging socket is marked as the horizontal and vertical coordinate points in this coordinate system, and the corresponding spatial coordinates are extracted and generated. An addressing mapping table between the spatial coordinates and the topology database is established, and the generated spatial coordinates are used as index keys to uniquely bind with the corresponding electrical end node ID in the construction wiring topology model. Subsequently, the construction wiring topology data of the charging station is retrieved to obtain the actual physical length of the cable from the secondary lead of the transformer in the distribution area along the power supply cable tray to the specific socket, and the physical wire diameter and material properties of the conductor segment are read simultaneously. Based on the physical impedance mapping relationship that the conductor impedance is directly proportional to the length and inversely proportional to the conductor cross-sectional area, combined with the standard resistivity parameters of the conductor material, the total impedance reference value of the line from the transformer to the socket is calculated and accumulated segment by segment, and this value is used as the feeder impedance corresponding to the socket.

[0028] By combining multimodal hardware devices such as distribution area edge gateways, visual sensing terminals, and high-frequency sampling circuits for collaborative data acquisition, millisecond-level waveforms are captured at the initial conduction moment and the physical wire impedance is quantified using a two-dimensional grid. This tightly binds the abstract data acquisition steps with specific physical sensing entities. It ensures the objectivity and authenticity of various load, electrical parameter, and spatial status data sources, providing high-confidence underlying physical data support for subsequent power scheduling.

[0029] Further, the process of calculating the global capacity limit and environmental limit includes: obtaining the distribution network transformer capacity and expected fluctuating load; calculating the difference between the transformer capacity and the sum of the substation load and the expected fluctuating load to generate a dynamic power supply margin; multiplying the inrush frequency, preset time window, and rated power to predict the tidal load increment; comparing the dynamic power supply margin with the tidal load increment to determine the global capacity limit; extracting the time-domain waveform features of the transient voltage to obtain the ohmic voltage drop amplitude and relaxation time constant; multiplying the ohmic voltage drop amplitude and the relaxation time constant to generate a polarization scalar; obtaining historical temperature sequences and historical transient voltages; dividing the historical temperature sequence into multiple temperature intervals according to a preset temperature step size; and extracting the historical transient voltage at each of the aforementioned temperatures. The average local pressure drop within the temperature range is calculated, and the temperature center point values ​​of each temperature range are spliced ​​with the corresponding average local pressure drop values ​​in two-dimensional space to construct an impedance temperature drift matrix. Absolute temperature data is extracted from the temperature range and used as input keys. A spline interpolation algorithm is then used to perform a continuous mapping solution on the impedance temperature drift matrix to obtain the corresponding temperature pressure drop mapping value. This mapping value is divided by the system's preset standard environmental reference pressure drop and normalized to generate a dimensionless thermodynamic impedance correction factor. The ratio of the rated power divided by the thermodynamic impedance correction factor and the polarization scalar product is calculated to obtain an initial quotient. This initial quotient is compared with the rated power, and the smaller of the two is established as the environmental limit for each socket. The specific process is as follows: Figure 2 As shown.

[0030] The calculation process for the global capacity limit first involves retrieving the rated capacity data of the transformers in the current access area from the distribution network dispatch system, and simultaneously obtaining the expected fluctuating load value of the current area within a preset short-term future period from the upper-level power grid. The obtained current distribution network area load is summed with the expected fluctuating load to obtain the estimated total basic grid load. Then, the transformer capacity is subtracted from this total basic grid load, and the difference is taken as the dynamic power supply margin that the area can additionally allocate to charging stations at the current moment. Next, a pre-set time window length is extracted, and the obtained vehicle entry frequency is multiplied by this time window length to obtain the estimated total number of arriving vehicles within this time period. This total number of vehicles is then multiplied by the rated power of a single charging socket to calculate the expected tidal load increment that will be superimposed on the power grid. Finally, the smaller value between the dynamic power supply margin and the tidal load increment is selected as the global capacity limit that the charging area can allow to output.

[0031] For the generation process of polarization scalar, time-domain waveform features are extracted from the acquired transient voltages of each socket. Specifically, in the captured voltage drop and recovery waveforms, the initial trigger point where the voltage curve undergoes an initial abrupt change is located. The absolute value of the instantaneous voltage difference between this initial trigger point and the lowest point of the vertical voltage drop is measured and extracted as the ohmic voltage drop amplitude. Subsequently, the voltage recovery curve after the lowest point is traced to locate the steady-state recovery voltage point where the recovery curve tends to flatten. The voltage difference between the lowest point and the steady-state recovery voltage point is calculated as the total recovery amplitude. Based on the physical characteristic proportion of the exponential step response of the first-order circuit (i.e.,...),... The target voltage characteristic threshold corresponding to the minimum point voltage plus approximately 63.2% of the total recovery amplitude is calculated. The characteristic time point at which the voltage first crosses the target voltage characteristic threshold is retrieved in ascending time sequence from the time-domain waveform data sequence. The time span between this characteristic time point and the time point corresponding to the minimum point is calculated and extracted as a relaxation time constant. After extraction, the ohmic voltage drop amplitude and the relaxation time constant are arithmetically multiplied to obtain the absolute value of the polarization characteristic that comprehensively characterizes the battery's ohmic internal resistance and diffusion dynamics. To eliminate dimensional differences, this absolute value of the polarization characteristic is divided by a preset healthy battery reference polarization constant (pre-calibrated by a new factory standard battery under normal testing) for normalization. The resulting dimensionless ratio is the polarization scalar of the corresponding socket (under typical operating conditions, the polarization scalar of a new battery is approximately 1.0-1.2, a moderately aged battery is approximately 1.5-2.0, and a severely aged battery is usually greater than 2.5).

[0032] To ensure that the extracted polarization scalar accurately and objectively reflects the electrochemical aging and polarization heating tendency of the electric bicycle's internal battery, and to eliminate transient interference from external circuit impedance and charger input capacitance, the above extraction process and its benchmark parameter calibration must meet the following test conditions and preprocessing mechanisms: First, regarding the hardware sampling node arrangement, the voltage sampling probe of the high-frequency sampling circuit built into the socket is physically positioned on the DC side after the power supply output terminal of the socket (if the charging station provides DC output) or close to the output terminal of the control relay, and the sampling frequency is set to no less than 10kHz to ensure accurate capture of millisecond-level voltage relaxation waveforms; second, the reference polarization constant of the healthy battery is not an arbitrarily specified empirical value, but is obtained in advance based on rigorous offline calibration experiments: a mainstream model of brand-new original battery pack adapted to the target charging station is selected, and the battery is tested in a standard constant temperature environment (25℃±2℃) and a standard initial state of charge (SOC of 50%). Under the rated trial charging current (e.g., constant current 2A), the same physical connection and transient conduction test of the socket are performed; the ohmic voltage drop amplitude and relaxation time constant under the standardized conditions are extracted and multiplied, and the arithmetic mean obtained from multiple tests is established as the reference polarization constant of the healthy battery, which is then fixed in the non-volatile memory of the edge computing gateway of the distribution area as a global benchmark; finally, when extracting the socket polarization scalar in actual operation, the system's underlying drive circuit adopts a constant current soft-start strategy (controlling the initial conduction current to a small constant current value consistent with the calibration experiment), and only the transient voltage waveform during the constant current soft-start stage is extracted for feature extraction.

[0033] To construct the impedance temperature drift matrix, the system retrieves historical temperature sequences and corresponding historical transient voltage data sets synchronously recorded within a preset historical period from the system storage medium. A fixed temperature value is set as the preset temperature step size. Using this step size as the interval, the global temperature range covered by the historical temperature sequence is equally divided into n continuous discrete temperature intervals. For each divided temperature interval, all historical transient voltage waveforms occurring within that temperature range are retrieved, and the voltage drop values ​​from these waveforms are extracted and introduced... The criteria remove abnormal high-frequency jitter values ​​and distortion flypoints that deviate from the distribution range. The arithmetic mean of the remaining effective pressure drop values ​​is calculated to generate a local pressure drop mean specific to that temperature range. Simultaneously, the midpoint between the upper and lower limits of that temperature range is determined as the temperature center point value. The temperature center point value of each temperature range is used as the x-axis, and the corresponding local pressure drop mean is used as the y-axis. Two-dimensional spatial data points are then stitched together to construct an impedance temperature drift matrix containing multiple sets of discrete mapping relationships. in, Represents the impedance temperature drift matrix; This represents the total number of discrete temperature intervals divided into regions. These represent the temperature center point values ​​for the 1st to nth temperature intervals, respectively. These represent the average local pressure drop values ​​for the 1st to nth temperature intervals corresponding to the aforementioned temperature center points.

[0034] For the process of establishing environmental limits, the current absolute temperature data is parsed from the collected charging area temperature data. This absolute temperature data is used as the input key for the target query. A spline interpolation algorithm is called to fit a smooth, continuous mapping curve between the discrete coordinate data points of the impedance temperature drift matrix. The absolute temperature data value is substituted into this continuous mapping curve for interpolation addressing to solve for the corresponding ordinate mapping value. The mapping value is divided by the system's preset standard environmental reference pressure drop to obtain a dimensionless thermodynamic impedance correction factor as the output. Finally, the generated thermodynamic impedance correction factor is multiplied by the extracted polarization scalar to generate a comprehensive attenuation denominator. To prevent the denominator from minimizing, which could lead to calculation overflow or infinite power output, the comprehensive attenuation denominator is compared with the system's preset minimum safety boundary base, and the larger value is taken as the final effective denominator. The rated power of the corresponding charging socket is divided by this effective denominator to calculate the initial quotient. This initial quotient is compared with the rated power, and the smaller value is taken as the environmental limit for the charging socket under the current thermodynamic and electrochemical environmental constraints. The calculation logic is expressed as follows: in, Indicates environmental limits; This indicates the rated power of the corresponding charging socket; This represents the thermodynamic impedance correction factor (dimensionless). Represents the polarization scalar (dimensionless).

[0035] By comparing the dynamic power supply margin with the tidal load increment and selecting the smaller value to determine the global limit, and using the product of the ohmic voltage drop amplitude and the relaxation time constant to characterize the polarization state, and combining the interpolation mapping of the impedance temperature drift matrix to establish the environmental limit, the impact of tidal power consumption on the power grid is not only quantified, but also related to the thermodynamic characteristics of the micro battery, thus enhancing the physical basis for the environmental derating control of the socket node.

[0036] Furthermore, the process of dividing the area into dual-feature region, single-feature region, and unconstrained region includes: extracting the full-load operating current of each socket under rated power; multiplying the full-load operating current with the corresponding feeder impedance to quantify and generate the theoretical maximum line loss voltage drop of each socket, which is used as the line voltage drop feature; using spatial coordinates to perform illumination tracing and shadow projection calculations in a preset three-dimensional building model of the station to extract the cumulative exposure time of each socket under direct sunlight, which is used as the sunlight exposure feature; and combining the line voltage drop feature with the sunlight exposure feature. A two-dimensional feature space is constructed, and extreme values ​​for voltage drop risk and exposure duration are set within this space. The set of sockets whose line voltage drop characteristics exceed the extreme value for both voltage drop risk and sunlight exposure characteristics exceed the extreme value for exposure duration is classified as a dual-feature region. The set of sockets whose only characteristics exceed the extreme value for either voltage drop risk or sunlight exposure characteristics exceed the extreme value for exposure duration is classified as a single-feature region. The set of sockets whose only characteristics exceed the extreme value for both voltage drop risk and sunlight exposure characteristics exceed the extreme value for exposure duration is classified as an unconstrained region. The specific process is as follows: Figure 3 As shown.

[0037] The quantitative generation process for line voltage drop characteristics first involves reading the rated power value of each socket and combining it with the nominal operating voltage of the power supply circuit where the charging network is located. The rated power is then divided by the nominal operating voltage to calculate the full-load operating current value of each socket under full-load operation. Subsequently, the extracted full-load operating current value is directly multiplied by the obtained feeder impedance value of the corresponding socket to calculate the absolute value of the maximum voltage loss that will occur on the power supply line under the theoretical maximum load transmission condition. This absolute value of the loss is then extracted as the line voltage drop characteristic of the corresponding socket.

[0038] To address the challenges of extracting sunlight exposure characteristics and adapt to the computing power limitations of the edge computing gateway in the substation area, a combination of offline pre-computation in the cloud and real-time table lookup at the edge is employed. In the high-performance computing node in the cloud, a pre-built 3D building model of the charging station, including its roof, supporting columns, and surrounding adjacent buildings, is retrieved, and the actual geographical latitude and longitude coordinates of the station are input into the system. Obtain the system timestamp of the actual date of the current scheduled task (usually with a daily or weekly pre-calculation update cycle), and use it as the epoch parameter for calculating the solar declination angle. Based on the local natural time, use the astronomical ephemeris algorithm to calculate the real-time altitude and azimuth angles of the sun from sunrise to sunset sequentially at a preset discrete time step. Use the spatial coordinates of each socket as the ray emission origin, construct the light source direction vector based on the solar altitude and azimuth angles corresponding to each time step, and emit a virtual test ray in the reverse direction of the light source direction vector. Perform spatial collision detection, use the bounding box hierarchical structure tree algorithm to filter irrelevant spatial regions, and call the Möller-Trumbore intersection algorithm to calculate the intersection points of the virtual test ray with the candidate geometric triangular facets in the model. If the virtual test ray intersects with any geometric mesh facet in the three-dimensional building model of the site before reaching the coordinates of the target socket, it is determined that the socket is in a shadow projection state within that time step; otherwise, if the test ray reaches the socket coordinates without obstruction, it is determined that it is in a direct sunlight state. During the offline calculation phase, the cloud accumulates and statistically analyzes all time steps across which each socket is determined to be under direct sunlight, using this as the sunlight exposure characteristic of the corresponding socket in the natural environment. The calculation results for all sockets are then solidified into a discrete feature lookup table and sent to the local edge computing gateway of the charging station. During the feature extraction phase of actual operation scheduling, the edge computing gateway directly retrieves the corresponding sunlight exposure characteristic value from the feature lookup table based on the current date and time and the socket ID, thereby eliminating the computational overhead of the large and redundant 3D geometry engine on the edge computing side.

[0039] The specific implementation process for regional clustering uses the obtained sunlight exposure characteristics as the horizontal axis variable and the line voltage drop characteristics as the vertical axis variable, mapping each socket to a two-dimensional feature coordinate point, and integrating them to construct a two-dimensional feature space for all sockets. Combining the allowable deviation standard of voltage drop in power grid operation and the critical time of environmental heat accumulation, the preset allowable deviation percentage of the rated voltage of the power supply network is extracted from the system configuration table, multiplied by the rated voltage to obtain the absolute voltage drop safety upper limit, thereby establishing the voltage drop risk extreme value; pre-stored battery casing thermal runaway exposure test parameters are retrieved to extract the minimum continuous illumination duration that triggers a dangerous temperature rise, thereby establishing the exposure duration extreme value; a horizontal voltage drop risk extreme value baseline and a vertical exposure duration extreme value baseline are drawn in the two-dimensional feature space. All feature coordinate points in the two-dimensional feature space are traversed and compared. When a characteristic coordinate point corresponding to a socket is detected, and the value of its line voltage drop characteristic exceeds the extreme value baseline of voltage drop risk, and at the same time, the value of its sunlight exposure characteristic also exceeds the extreme value baseline of exposure duration, it is determined that the node faces the danger of deep electrothermal coupling, and it is clustered into the dual feature region set. When a characteristic coordinate point corresponding to a socket is detected, if only the line voltage drop characteristic value exceeds the voltage drop risk extreme value baseline, or only the sunlight exposure characteristic value exceeds the exposure duration extreme value baseline, it is determined that the node faces a unilateral constraint boundary and is clustered into a single feature region set. When a characteristic coordinate point corresponding to a socket is detected, and its line voltage drop characteristic value does not exceed the voltage drop risk extreme value baseline, and its sunlight exposure characteristic value does not exceed the exposure duration extreme value baseline, the node is determined to be in a normal safe state, and it is clustered into the unconstrained region set.

[0040] The line voltage drop is generated by deriving the product of the full-load operating current and the feeder impedance based on Ohm's law. Three-dimensional illumination tracing is used to extract the cumulative exposure time as a solar exposure feature. Boolean logic thresholds are used to classify complex charging stations into dual-feature, single-feature, and unconstrained zones. The electrical properties of nodes are objectively and physically quantified and integrated with the spatial photothermal environment, enabling rapid and structured identification of potential risk nodes within the area.

[0041] Further, the process of generating updated limits includes: traversing all sockets in each partition and marking the socket with the highest polarization scalar value as the reference socket; calculating the Euclidean distance between each socket and the reference socket based on the spatial coordinates; filtering out sockets whose Euclidean distance is less than the preset distance threshold to construct the adjacent set; dividing the Euclidean distance corresponding to each socket in the adjacent set by the preset distance threshold to generate a spatial thermal attenuation coefficient less than 1; multiplying the original environmental limit of each socket in the adjacent set by the corresponding spatial thermal attenuation coefficient to calculate and generate the derated updated limit.

[0042] For the marking process of the reference socket, the system performs an independent traversal operation according to the aforementioned partitions. Within a single physical partition, the polarization scalar value of all charging sockets in that partition at the current moment is extracted, and this value is used as the sorting key to perform a descending order sorting operation. The charging socket located at the first position in the descending sequence, i.e., the charging socket with the most intense polarization heat generation, is selected, and its physical spatial location is determined as the highest heat dissipation source in that partition, and it is assigned a unique identification mark as the reference socket. For the construction process of the adjacent set, the 3D Cartesian space coordinate data of the reference socket and the other sockets in the same partition are retrieved. The linear geometric span between the coordinate points of the other sockets and the coordinate points of the reference socket is calculated one by one, thereby obtaining the Euclidean distance value of the corresponding socket. The formula for calculating the 3D Euclidean distance is: in, This represents the Euclidean distance value corresponding to socket i. This represents the spatial coordinates of the reference jack in a 3D Cartesian coordinate system; This represents the spatial coordinates of the i-th socket visited. Subsequently, a preset distance threshold characterizing the thermal radiation safety boundary is introduced. This preset distance threshold is a critical physical distance calibrated through a full-load heating experiment (typically ranging from 0.5 meters to 1.5 meters). Tests have shown that when the socket spacing reaches or exceeds this critical physical distance, the local thermal radiation interaction between adjacent devices has attenuated to the level of background ambient noise and can be ignored. Therefore, the calculated Euclidean distance values ​​for each socket are... The values ​​are compared with the preset distance threshold, and edge socket nodes with Euclidean distance values ​​greater than or equal to the preset distance threshold are removed; the remaining socket nodes with Euclidean distance values ​​strictly less than the preset distance threshold are classified and extracted, and a thermal radiation cross-influence circle with the reference socket as the absolute center is established. All socket nodes within the influence circle are packaged to construct the adjacent set.

[0043] Regarding the process of generating update limits, for each charging socket in the adjacent set, its spatial identity attribute is determined; if the Euclidean distance value of the corresponding socket is detected to be 0, it is determined that the socket is the reference socket itself, and its update limit is directly forced to be set to 0 to perform absolute physical power-off isolation; if the Euclidean distance value of the corresponding socket is detected to be greater than 0, the exclusive Euclidean distance value is used as the dividend, and the preset distance threshold is used as the divisor to perform division operation. Although heat radiation attenuation in free space follows an inverse square law, within the small spatial scale of densely packed centralized charging stations (i.e., within the range of the preset distance threshold), heat transfer is affected by a combination of factors, including equipment casing shielding, air convection, and cable heat conduction. Engineering measurements have verified that the attenuation curve of this combined thermal effect is approximately linear within this small physical range. Therefore, using a linear distance ratio to simplify the calculation not only keeps the error within the acceptable safety margin for engineering but also effectively avoids the edge-side computational power consumption caused by square operations. Due to the distance condition setting of nodes within the set, the dividend is always strictly less than the divisor. This division operation will yield a continuous proportionality constant greater than 0 and strictly less than 1, which is defined as the spatial heat attenuation coefficient of the corresponding socket. Finally, the original environmental limit of the socket is extracted, and this original environmental limit is directly multiplied by the newly generated exclusive spatial heat attenuation coefficient. Utilizing the proportional reduction effect brought about by the product operation, the upper limit of the allowable charging power of the target socket is forcibly lowered. To avoid the power after derating falling into the startup dead zone of the electrical equipment and causing high-frequency oscillation of the relay, the preliminary absolute value obtained by the product calculation is compared with the minimum maintenance power limit of the single port preset by the system. If it is lower than the maintenance power limit, the limit is set to 0 directly; otherwise, the absolute value is retained and finally set as the updated limit of the socket after spatial physical thermal isolation derating.

[0044] By extracting the reference socket for polarization scalar ranking and calculating its Euclidean distance from surrounding sockets to generate a thermal attenuation coefficient, a thermal isolation protection mechanism based on spatial geometric correlation is established by proportionally reducing the limit of adjacent sets according to the distance. This reflects the physical attenuation law of heat radiation and accumulation between adjacent charging devices, improving the local environmental safety of the site.

[0045] Furthermore, the process of splitting the global capacity limit into power quotas corresponding to each zone includes: accumulating the updated limits and environmental limits of all sockets within the target charging area to generate a sum of limits for each socket; when the sum of limits for each socket is greater than the global capacity limit, for the dual-feature region, applying a first-level thermal impedance attenuation and voltage drop compensation pre-deduction to the global capacity limit, extracting the corresponding capacity to generate a first stringent quota; for the single-feature region, extracting the capacity corresponding to the current region and implementing a unilateral derating constraint to generate a second restricted quota; for the unconstrained region, subtracting the first stringent quota and the second restricted quota from the global capacity limit to obtain the remaining capacity limit, assigning it to the current region as a third regular quota.

[0046] First, all socket nodes within the target charging area are traversed. During this process, the updated limit values ​​of all sockets included in the adjacent set are extracted based on spatial thermal decay calculations. Simultaneously, the original environmental limit values ​​of all other regular sockets not in the adjacent set are extracted. The effective power limit values ​​of all sockets in the current area are summed to obtain the total limit value of all sockets in the target charging area. Then, the calculated total limit value of all sockets is compared with the global capacity limit value.

[0047] When the comparison result indicates that the sum of the limits for all sockets exceeds the global capacity limit, a tiered capacity quota splitting mechanism for different characteristic regions is initiated. Before quota splitting, constraint factors are pre-generated: the maximum value of the actual sunlight exposure characteristic duration and the maximum value of the actual line voltage drop characteristic amplitude of all over-limit sockets within each characteristic region are extracted as the worst representative value for the corresponding region; the relevant characteristic extreme values ​​generated for each socket are called, and the environmental heat exposure attenuation factor is calculated by dividing the set exposure duration extreme value by the actual sunlight exposure characteristic duration in the worst representative value. (in The voltage drop compensation pre-deduction factor is calculated by dividing the set voltage drop risk extreme value by the actual line voltage drop characteristic amplitude in the worst-case representative value. (in ).

[0048] First, the dual-feature region is processed by extracting the sum of the original limit values ​​of all sockets in the current region as its initial pre-allocated capacity. The sum of the initial limit values ​​of all sockets in the entire region is then obtained. The global capacity limit is extracted according to the proportion of the pre-allocated capacity of this dual-feature region, and multiplied successively by the environmental heat exposure attenuation factor and the voltage drop compensation pre-deduction factor. The first-level thermal impedance attenuation and voltage drop compensation pre-deduction are implemented using the following formula. The calculated absolute capacity value generates the first stringent quota assigned to this dual-feature region: in, This indicates the first stringent quota assigned to the dual-feature region; Indicates the global capacity limit; This represents the initial pre-allocated capacity of the dual feature regions; This represents the sum of the limit values ​​for all sockets in the target charging area obtained from the preceding steps. Indicates the environmental heat exposure attenuation factor; This indicates the pre-deduction factor for pressure drop compensation.

[0049] After processing the dual-feature area, quota reduction processing is performed for the single-feature area. The sockets within the single-feature area are further identified and divided into a first attribute subset (only sunlight exposure exceeding limits) and a second attribute subset (only line voltage drop exceeding limits). The cumulative value of all socket limits within the first attribute subset is extracted as the first initial demand capacity, and the cumulative value of all socket limits within the second attribute subset is extracted as the second initial demand capacity. The global capacity limit is extracted as a basic share according to the proportion of the demand capacity of the two subsets. For the first attribute subset, the extracted basic share is used to perform a unilateral reduction by multiplying the single environmental heat exposure attenuation factor; for the second attribute subset, the extracted basic share is used to perform a unilateral reduction by multiplying the single voltage drop compensation pre-deduction factor. The final capacity values ​​obtained after the above two unilateral reduction operations are accumulated to generate the second restricted quota assigned to the single-feature area. The calculation logic is expressed as follows: in, This indicates a second restricted quota assigned to the single feature region; This represents the sum of all socket limits within the first attribute subset, i.e., the first initial demand capacity; This represents the sum of all socket limits within the second attribute subset, i.e., the second initial demand capacity; This represents the sum of the limit values ​​for all sockets in the target charging area obtained from the preceding steps. Indicates the global capacity limit; Indicates the environmental heat exposure attenuation factor; This indicates the pre-deduction factor for pressure drop compensation.

[0050] Finally, for the unconstrained zone, the remaining quota is cleared and allocated. The total number of global capacity limits in the system is retrieved as the minuend, and the first stringent quota value and the second restricted quota value generated in the previous process are read as the subtrahends. An arithmetic subtraction operation is performed, and the first stringent quota and the second restricted quota are subtracted from the global capacity limit to obtain the subtraction difference representing the remaining available resources. The remaining capacity limit indicated by the difference is fully extracted and directly assigned to the unconstrained zone, which is established as the third regular quota for the current region.

[0051] By implementing a tiered capacity pre-deduction and unilateral derating strategy for dual-characteristic regions, single-characteristic regions, and unconstrained regions when the total limit is exceeded, the total capacity is progressively divided into a first stringent quota, a second restricted quota, and a third regular quota from top to bottom. While prioritizing the restriction of node power output under harsh physical conditions, the remaining capacity in safe regions is fully allocated and allocated, ensuring both the system's electrical safety boundaries and maintaining charging service efficiency.

[0052] Furthermore, the evaluation model includes: a topology reconstruction layer, which maps and reconstructs all charging sockets into a virtual power supply topology tree with hierarchical parent-child relationships based on the branch connection relationships in the construction wiring topology data and the feeder impedance of each charging socket, and uses the polarization scalar as the attribute feature of the corresponding node; and a line-side environment interaction layer, which classifies each target socket in the virtual power supply topology tree as a core node, and classifies the remaining charging sockets that share the same upstream power supply path as homogeneous environment nodes; and counts the number of homogeneous environment nodes whose polarization scalar is greater than the specified value. The number of nodes in the core node polarization scalar is calculated, and the ratio of this number to the total number of nodes in the same source environment is used to generate a path congestion feature value. The collaborative decision layer concatenates the polarization scalar of the target socket, its depth level in the virtual power supply topology tree, and the path congestion feature value into a context feature vector, which is then input into the classification network structure to output the corresponding scheduling feature value. Wherein, when the depth level is greater than a preset depth level and the path congestion feature value is greater than a preset congestion extreme value, the classification network structure is forced to output the scheduling feature value ranked last.

[0053] For the topology reconstruction layer of the evaluation model, the feeder impedance and polarization scalar data of all charging sockets within the target charging area are first obtained. The power supply starting point of the distribution network transformer is set as the root node of the virtual power supply topology tree. Based on the branch connection relationships in the construction wiring topology data and the feeder impedance values ​​of each charging socket, a physical power supply hierarchy mapping logic is established. Nodes with smaller feeder impedances are mapped to upstream parent nodes closer to the root node, and nodes sharing the same power supply line and with larger feeder impedances are mapped to their corresponding downstream child nodes. This establishes the hierarchical parent-child relationship of all charging sockets in the topology tree. After the topology tree is constructed, the polarization scalar value corresponding to each socket is directly assigned to the corresponding node position in the topology tree as an inherent attribute feature of that node. Simultaneously, the tree structure is traversed forward, and the total number of topology connections crossing each node from the root node is counted. This total number of crossings is quantified and extracted as the depth level of the corresponding socket in the virtual power supply topology tree.

[0054] For the interaction layer along the evaluation model, a polling mechanism is used to traverse each charging socket in the virtual power supply topology tree. When a target socket is reached, its physical state is marked as the core node in the current calculation cycle. Then, starting from this core node, a reverse backtracking addressing is performed in the topology tree until the root node, extracting the complete upstream power supply path that the core node depends on. Next, a structural search is performed in the global topology tree to filter out all other charging sockets whose physical access points are also located on the same upstream power supply path segment. These are uniformly classified as homologous environment nodes of the core node, and the total number of homologous environment nodes is calculated. The polarization scalar of the core node itself is extracted, and the polarization scalars of all homologous environment nodes are read one by one for numerical comparison. The number of homologous environment nodes whose polarization scalar values ​​are strictly greater than the polarization scalar value of the core node is recorded. Finally, determine whether the total number of nodes in the same environment is 0. If it is 0, the path congestion characteristic value of the core node is directly determined to be 0. If it is not 0, the number of nodes that are greater than the polarization scalar of the core node is used as the dividend, and the total number of nodes in the same environment is used as the divisor to perform a division operation. The resulting numerical ratio is defined as the path congestion characteristic value corresponding to the core node.

[0055] For the collaborative decision-making layer of the evaluation model, the polarization scalar value of the current target socket, its depth level value established in the virtual power supply topology tree, and the derived path congestion feature value are extracted. Following a preset data alignment order, these three dimensions are concatenated one-dimensionally to generate a context feature vector for the target socket. Before inference, the classification network structure is pre-constructed and trained: a multilayer perceptron network containing an input layer, multiple hidden layers, and an output layer is constructed; historical charging operation records are obtained, and the polarization scalar value, depth level value, and path congestion feature value of historical sockets are extracted and concatenated to form a historical context feature vector sample. The maximum operating power level when no abnormal alarm occurred at the corresponding node in the historical record is extracted as the true classification label of the sample; the historical context feature vector sample is input into the multilayer perceptron network, and the predicted classification probability is output; the error between the predicted classification probability and the true classification label is calculated using the cross-entropy loss function, and the weight parameters in the multilayer perceptron network are updated through backpropagation using the gradient descent algorithm until the error converges to a preset range, completing the training of the classification network structure. The context feature vector is used as the sole input data and fed forward into a pre-configured classification network structure. After numerical mapping and activation in the hidden layers, the corresponding classification result is obtained from the network's output. This classification result is directly mapped to a specific level in the system's preset discrete power level table, defining it as the initial scheduling feature value.

[0056] Before outputting the final scheduling characteristic value, a logical bypass determination is performed by retrieving the preset level depth threshold and preset congestion extreme value configured in the system. The preset level depth threshold is the maximum safe topology level calculated by rounding down the ratio of the maximum allowable voltage drop limit at the end of the distribution network to the average impedance voltage drop of a single feeder segment. The preset congestion extreme value is the critical upper limit threshold for the proportion of highly polarized equipment clustered within the same branch, defined based on historical thermal runaway accident statistics. The depth level value of the current target socket is compared with the preset level depth threshold, and simultaneously, the path congestion characteristic value of the target socket is compared with the preset congestion extreme value. When the logical determination result simultaneously satisfies that the depth level is greater than the preset level depth threshold and the path congestion characteristic value is greater than the preset congestion extreme value, a system-level forced overwrite action is triggered. The system directly intercepts and discards the initial scheduling feature value output by the classification network structure, and forcibly extracts the scheduling feature value that is at the end of the preset discrete power level table (i.e., representing the lowest output power or physical disconnection state) as the final output of the target socket; if the above dual threshold conditions are not met, the bypass interception is lifted, and the initial scheduling feature value output by the classification network structure is directly output as the final scheduling feature value.

[0057] Before generating the underlying control data, the partition quota constraint solution is performed: the expected output power corresponding to the current candidate scheduling feature values ​​of all sockets in the same partition is summed to generate the sum of the partition; the sum is compared with the power quota corresponding to the partition; if the sum is greater than the power quota of the partition, the candidate scheduling feature values ​​of the corresponding sockets are forced to be lowered to a lower power level in order of increasing polarization scalar of each socket, and the sum is recalculated until the sum is not greater than the power quota; at this time, the candidate scheduling feature values ​​of each socket are established as the final scheduling feature values.

[0058] First, a raw dataset for model training is constructed. For the initial deployment phase with no historical data (cold start), simulated charging and discharging data is generated by introducing a physical simulation model of the charging station. Combined with expert-calibrated safety boundary parameters, an initial pre-trained feature vector set is constructed. The mean and standard deviation of this pre-trained set are calculated to initialize the Z-score standardization algorithm. For the regular operation phase, the historical operation database of the charging station is directly accessed, and historical charging records are extracted according to a preset time span. From each record (including initial simulation records or historical real charging records), the polarization scalar value of the corresponding charging socket when charging occurs, the depth level value in the virtual power supply topology tree, and the path congestion feature value are extracted. These three dimensions are concatenated into an original one-dimensional feature vector, which is then used to introduce a normal distribution-based algorithm. An outlier detection algorithm removes distorted feature vectors caused by sensor communication failures or high-frequency electromagnetic interference, and obtains a cleaned set of effective feature vectors.

[0059] Since the physical dimensions and numerical distribution ranges of polarization scalar, depth level and path congestion feature values ​​are significantly different, in order to prevent gradient vanishing or local optima during network training, the Z-score normalization algorithm is used to perform a linear transformation on the effective feature vector set. The historical mean and standard deviation of each feature dimension in the effective feature vector set are calculated. Each feature value is subtracted from the historical mean of the corresponding dimension and then divided by the standard deviation of the corresponding dimension to generate a normalized context feature vector with a mean of 0 and a variance of 1, which is used as the standard input sample of the model.

[0060] Retrieve the historical electrical monitoring logs corresponding to each valid feature vector to trace the highest temperature rise and maximum line loss voltage drop data of the charging socket throughout the entire charging cycle. Based on the preset safe temperature rise and safe voltage drop extreme values, backtrack and deduce the maximum allowable power level at which the socket can maintain safe operation under the current operating conditions. Map this maximum allowable power level to the system's preset discrete power level table (e.g., full power level, derating level, ultra-low power level, disconnect level), and convert the determined discrete level into a one-hot code vector as the true classification label for the corresponding standard input sample.

[0061] The standard input samples and the true classification labels are combined to form a complete supervised learning dataset. To ensure the model's generalization ability, the total number of samples in this supervised learning dataset is set to be at least tens of thousands. Simultaneously, to address the class imbalance problem caused by the high proportion of samples in normal safe operating states and the scarcity of samples in abnormal restricted states (such as reduced load or disconnection), the SMOTE oversampling algorithm is used to artificially synthesize and interpolate the minority class (abnormal restricted state) features during dataset construction, or a class weighting strategy is introduced in the subsequent loss function calculation to give the minority class a higher cross-entropy penalty weight. Subsequently, a randomization mechanism is introduced, dividing the complete supervised learning dataset into a training set for updating network parameters, a validation set for monitoring training and preventing overfitting, and a test set for finally evaluating generalization ability, according to a preset ratio (e.g., 80% training set, 10% validation set, 10% test set).

[0062] A multilayer perceptron is constructed as the classification network structure. In feature selection, polarization scalar represents the electrochemical heating tendency of micro-cell aging, depth level represents the inherent static line loss voltage drop of the power supply line, and path congestion feature value represents the local aggregated voltage drop and heat superposition caused by dynamic concurrent charging of the same branch. The combination of these three features constitutes a complete physical representation of the coupling risk of local overload and thermal runaway at the bottom layer of the charging station. To accurately fit the nonlinear mapping relationship between these three-dimensional physical features, the structure sequentially includes: an input layer with 3 nodes (corresponding to the 3 dimensions of the context feature vector), two hidden layers using the ReLU activation function (the first hidden layer is preset to have 64 neurons, and the second hidden layer is preset to have 32 neurons, used for progressive extraction of nonlinear cross features of multidimensional physical quantities), and an output layer with the number of nodes equal to the preset total number of discrete power levels in the system. The output layer uses the Softmax activation function to output the predicted probability distribution of each discrete power level, and the Xavier initialization algorithm is used to assign initial random values ​​to all connection weights and bias parameters in the multilayer perceptron network.

[0063] The pre-defined training set is input into the classification network structure in batches according to a preset batch size. Standard input samples are sequentially processed through linear multiplication and addition and non-linear activation mapping in the input layer and hidden layer. Finally, the Softmax function of the output layer outputs the predicted classification probability vector corresponding to each power level. The cross-entropy loss function is called to calculate the difference between the predicted classification probability vector and the corresponding true classification label (one-hot code vector), quantifying the classification error.

[0064] The Adam optimization algorithm is adopted. Based on the calculated classification error, the gradient of the error with respect to each connection weight and bias is calculated layer by layer in reverse using the chain rule. According to the set initial learning rate and the momentum information of the current gradient, all weights and bias parameters in the multilayer perceptron network are finely adjusted and updated in the opposite direction of the gradient.

[0065] Repeat the forward and backward propagation iterations on the training set. After each iteration, input the validation set into the current network to calculate the validation loss value. Continuously monitor the trend of the validation loss value. When it is detected that the training set loss value continues to decrease, but the validation set loss value no longer decreases or rebounds for several consecutive iterations, the network is determined to trigger an early stopping mechanism to prevent overfitting.

[0066] By reconstructing discrete charging sockets into a hierarchical virtual power supply topology tree with parent-child relationships, and comparing the polarization states of core nodes and nodes in the same source environment to calculate path congestion features, the model forces the output of the last feature value at the end of the power supply under high congestion conditions. This model structure extracts the superposition of heat accumulation and voltage drop on the same power supply branch, avoiding the uncontrollable operational risks brought about by simply relying on conventional neural networks.

[0067] Further, the process of controlling the contactor to disconnect when the degradation characteristic value is greater than the preset degradation extreme value includes: mapping the scheduling characteristic value to a preset discrete power level table in descending order of value, querying and generating the control data corresponding to each socket; outputting a drive level based on the control data, controlling the working circuit relay array built into the socket to close to the physical conduction path of the corresponding level; after the physical conduction path is closed, starting a timer and continuing for a preset stable delay period, reading the voltage difference between the input and output terminals of the working circuit relay array at the end of the preset stable delay period as the steady-state voltage drop, and simultaneously collecting the actual working current on the physical conduction path; calculating the current contact resistance by dividing the steady-state voltage drop by the actual working current, calculating the absolute difference between the current contact resistance and the preset nominal contact resistance, and using the absolute difference as the degradation characteristic value characterizing the internal resistance state of mechanical contact erosion; when the degradation characteristic value is greater than the preset degradation extreme value, generating a hardware-level physical isolation instruction, and using the hardware-level physical isolation instruction to directly drive the main circuit contactor connected in series with the working circuit relay array to perform a mechanical disconnection action.

[0068] First, the scheduling feature values ​​corresponding to each charging socket output in the previous steps are extracted. Before executing the scheduling action, the discrete power level table needs to be constructed in advance. The construction process is as follows: obtain the physical topology and adjustable impedance parameters of the relay array of the built-in working circuit of the charging socket, and calibrate multiple hardware-level physical output levels, including full-load through-pass, derating output, and physical disconnection levels; obtain the global numerical distribution space of the scheduling feature values ​​output by the aforementioned evaluation model, and divide the global numerical distribution space into multiple continuous feature value intervals according to the electrical safety threshold ladder preset by the system; bind the feature value intervals one-to-one with the levels of the physical output levels that allow output power from large to small in order of increasing danger, and write the mapping relationship into the non-volatile memory of the system to solidify and generate the discrete power level table.

[0069] When generating control commands, the discrete power level table is retrieved, and the scheduling feature values ​​of each extracted socket are mapped one by one to the specific feature value range in the discrete power level table. The target level control data that each socket should execute at the moment is queried and generated. Then, according to the generated control data, the corresponding high and low level signals are output to the outside through the drive circuit. The level signals are used to directly excite the electromagnetic coil of the working circuit relay array built into the corresponding socket, drive the corresponding mechanical contacts to perform a closing action, thereby closing the physical conduction path corresponding to the target level.

[0070] Taking the moment the drive level signal is output as the starting point, the inherent mechanical engagement time constant of the target relay is superimposed to determine that the physical conduction path has completed its initial closing action, and a timing mechanism is immediately initiated to continue for a preset stable delay period. The span of this preset stable delay period is designed to avoid the physical bounce and transient arc interference stage caused by the instant the mechanical contacts close; its specific value is set to be strictly greater than the maximum contact bounce time parameter specified in the target relay's manufacturer's specifications. When the timing reaches the end of the preset stable delay period, the voltage sensing acquisition circuits distributed at the input and output terminals of the working circuit relay array are triggered to perform synchronous voltage sampling. The voltage sensing acquisition circuits are controlled to continuously sample within at least one complete AC power frequency cycle to obtain the AC voltage time sequence at the input and output terminals. The effective voltage values ​​of the input and output AC voltage time series are calculated separately. An arithmetic subtraction operation is performed between the calculated input and output effective voltage values, and the difference is recorded as the steady-state voltage drop of the current closed circuit. The current sensing circuit is then controlled to synchronously acquire the actual effective value of the operating current on this physically conducting path. The preset nominal contact resistance value of the corresponding position of the working circuit relay array in its factory-new state is obtained from the system memory. The current contact resistance is calculated by dividing the actual acquired steady-state voltage drop by the actual effective value of the operating current. Then, the calculated current contact resistance is subtracted from the corresponding preset nominal contact resistance to obtain the absolute difference. This absolute difference reflects the actual increase in contact resistance caused by metal oxidation and surface pitting due to long-term switching, and is established as the degradation characteristic value. Finally, the preset degradation extreme value set by the system is retrieved. This preset degradation extreme value is the critical voltage drop value corresponding to the increase in contact resistance calculated based on the maximum allowable safe heating power and full-load rated operating current of the working circuit relay. The established degradation characteristic value is compared with the preset degradation extreme value. When the comparison result shows that the degradation characteristic value is greater than the preset degradation extreme value, a low-level hardware-level physical isolation command is generated. This hardware-level physical isolation command is directly transmitted to the main circuit contactor connected in series with the working circuit relay array, driving the tripping mechanism of the main circuit contactor to perform a mechanical disconnection action, completely severing the overall physical power supply connection of the socket.

[0071] As a preferred hardware implementation, the working circuit relay array specifically includes a full-load direct-through branch, a derating current control branch, and an ultra-low power branch arranged in parallel. Wherein: The full-load direct-through branch has a built-in first relay, which is a low-impedance direct-conduction circuit; the derating current control branch has a built-in second relay and a first current-limiting impedance component (preferably a high-power low-resistance current-limiting resistor or reactor) connected in series with the second relay; the ultra-low power branch has a built-in third relay and a second current-limiting impedance component connected in series with the third relay, and the equivalent resistance of the second current-limiting impedance component is strictly greater than the equivalent resistance of the first current-limiting impedance component.

[0072] When the control data indicates that the current target gear is the full-load direct-through gear, the drive circuit outputs a first combination control level, driving the first relay to close while keeping the second and third relays open. At this time, the charging input current flows completely through the full-load direct-through branch, achieving lossless full-power output. When the control data indicates that the current target gear is the derating output gear, the drive circuit outputs a second combination control level, driving the second relay to close while keeping the first and third relays open. At this time, the charging input current is forced to flow entirely through the first current-limiting impedance component, utilizing its hardware impedance... This generates a voltage reduction and current control effect, achieving the first stage of derating power output. When the control data indicates that the current target level is the extremely low power level, the drive circuit outputs a third combination control level, driving the third relay to close and keeping the first and second relays in the open state. At this time, the charging input current flows through the second current-limiting impedance component with a larger resistance, achieving a deeper level of hardware-level current-limiting safety output. When the control data indicates that the current target level is the off position, the drive circuit controls the first, second, and third relays to perform a complete disconnection action, so that the working circuit of the socket completely loses its physical electrical path.

[0073] By setting a preset stable delay period after the control relay closes, transient bouncing arcs are avoided to accurately read the steady-state voltage drop. The difference between this steady-state voltage drop and the nominal voltage drop is used to quantify the degree of erosion and deterioration of the mechanical contacts, driving the main circuit contactor to disconnect when the limit is exceeded. A two-level hardware physical isolation response system is established, forming a closed loop between algorithm control and actual electrical hardware status feedback, effectively preventing the potential for electrical overheating caused by contact aging.

[0074] Example 2: First, the edge computing gateway of the distribution network acquires the load of the distribution network area currently operating at a high level, and the visual sensing terminal continuously captures and accumulates vehicle arrival events to generate a high-level entry frequency. At the same time, the environmental sensing node collects the temperature of the area currently in a high-temperature state, and the high-frequency sampling circuit captures the transient voltage waveforms of each electric bicycle when multiple electric bicycles are inserted at the same time. The system background synchronously maps and extracts the feeder impedance and spatial coordinates of these sockets.

[0075] Next, due to the high load in the transformer area and the anticipated large increase in tidal load, the system calculates the extremely scarce dynamic power supply margin at the current moment and locks in a low global capacity limit accordingly. For each charging socket, the system extracts the ohmic voltage drop amplitude and relaxation time constant to generate a polarization scalar, identifies that some vehicles equipped with aging batteries have a serious tendency for polarization heating, and, in conjunction with the high-temperature environment, retrieves the impedance temperature drift matrix to generate initial environmental limits for each socket.

[0076] Subsequently, for charging sockets located at the end of the charging station (where the feeder impedance is high, resulting in a high line voltage drop) and on the edge of western exposure (where the sunlight exposure characteristic indicates prolonged exposure), the system classifies them into dual-characteristic zones; for sockets located deep inside the charging shed that are not exposed to sunlight but are at the end of the power supply, or sockets near the transformer but exposed to sunlight, they are classified into single-characteristic zones; and for sockets that are both near the transformer and in a shady place, they are classified into unconstrained zones.

[0077] Within the dual-feature region, the system accurately locates the vehicle socket with the highest polarization scalar and the most severe heat generation as the reference socket. Subsequently, using this reference socket as the center, the system groups nearby sockets into an adjacent set. To prevent a cascading effect of thermal runaway, the system directly and forcibly lowers the environmental limits of the remaining sockets in this adjacent set based on the distance attenuation ratio, generating updated limits.

[0078] During the quota allocation phase, due to the huge concurrent charging demand during the evening peak, the sum of the limits for each socket is far greater than the current scarce global capacity limit. The system triggers a tiered quota allocation mechanism: priority is given to the dual-characteristic area with the most severe environment, which is subject to double attenuation deduction of thermal resistance and voltage drop, and is only given the first stringent quota to maintain extremely low power operation; a unilateral derating is implemented for the single-characteristic area, which is given the second restricted quota; finally, the remaining global capacity limit is given to the unconstrained area as the third regular quota, so as to maximize the charging efficiency of the safe area while ensuring safety.

[0079] Finally, under the hard constraint of the aforementioned quota, the system reconstructs a virtual power supply topology tree and calculates high-level path congestion feature values ​​for branches with localized clustered charging. The system concatenates the relevant features and inputs them into the already trained classification network structure. Because some end sockets are located deep in the topology and are severely congested, the system triggers a logical bypass, forcibly outputting the scheduling feature value of the last ranked socket.

[0080] The underlying drive circuit controls the working circuit relay to close to the corresponding derating or disconnect position based on the scheduling characteristic value. After avoiding the preset stabilization delay period of contact bounce, the system collects the steady-state voltage drop at this time and compares it with the factory nominal voltage drop. In this scenario, the system detects a socket that has issued a derating command. Its mechanical contacts have developed severe ablation resistance due to long-term high-temperature and high-current operation, and the calculated degradation characteristic value exceeds the preset degradation limit. The system immediately generates a hardware-level physical isolation command, directly driving the main circuit contactor connected in series with the relay to perform a mechanical disconnect action, cutting off the physical power supply to the aging socket.

[0081] 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 scheduling electric bicycle charging based on load prediction, characterized in that, include: The system acquires the distribution network's substation load, entry frequency, and rated power of each socket; it acquires the charging area temperature, transient voltage of each socket, feeder impedance, and spatial coordinates; based on the substation load, entry frequency, and rated power, it calculates the global capacity limit; it extracts the time-domain waveform features of the transient voltage, acquires the ohmic voltage drop amplitude and relaxation time constant, and multiplies the ohmic voltage drop amplitude and relaxation time constant to generate a polarization scalar; it acquires historical temperature sequences and historical transient voltages, dividing the historical temperature sequences into multiple temperature intervals according to a preset temperature step size; it extracts the local voltage drop mean of the historical transient voltage in each temperature interval, and spline interpolates the temperature center point values ​​of each temperature interval with the corresponding local voltage drop mean in two-dimensional space to construct an impedance temperature drift matrix; it extracts the absolute temperature data, uses the absolute temperature data as input key values, and uses a spline interpolation algorithm to perform continuous mapping in the impedance temperature drift matrix to obtain the corresponding temperature voltage drop mapping value; it divides the mapping value by the system's preset standard environmental reference voltage drop, performs normalization processing, and generates a dimensionless thermodynamic impedance correction factor; The initial quotient is obtained by dividing the rated power by the product of the thermodynamic impedance correction factor and the polarization scalar. This initial quotient is then compared with the rated power, and the smaller of the two is taken as the environmental limit for each socket. Based on feeder impedance and spatial coordinates, line voltage drop and solar exposure characteristics are extracted and divided into regions: dual-characteristic region, single-characteristic region, and unconstrained region. Within each region, the socket with the largest polarization scalar is extracted as the reference socket. Based on spatial coordinates, sockets whose distance from the reference socket is less than a preset distance threshold are selected to form an adjacent set. The environmental limits of the adjacent set are updated using the ratio of distance to the distance threshold to generate updated limits. When the sum of the limits of all sockets is greater than the global capacity limit, the global capacity limit is split into power quotas corresponding to each region. Under power quota constraints, the polarization scalar and the depth level and path congestion feature values ​​extracted from the topology tree constructed based on the feeder impedance are input into the evaluation model to output scheduling feature values, and control data is generated to control the relay switching position. The feeder impedance topology tree is constructed by mapping and reconstructing all charging sockets into a virtual power supply topology tree with hierarchical parent-child relationships based on the branch connection relationship in the construction wiring topology data and the feeder impedance of each charging socket, and the polarization scalar is used as the attribute feature of the corresponding node. The path congestion feature is defined as follows: for each target socket in the virtual power supply topology tree, it is classified as a core node, and the other charging sockets that share the same upstream power supply path are classified as homogeneous environment nodes. The number of nodes in the same source environment with a polarization scalar greater than that of the core node is counted, and the ratio of this number to the total number of nodes in the same source environment is calculated to generate path congestion feature values. Collect the steady-state voltage drop after the relay is closed; generate degradation characteristic values ​​using the difference between the steady-state voltage drop and the preset nominal voltage drop; The contactor is disconnected when the degradation characteristic value is greater than the preset degradation extreme value.

2. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The process of acquiring the distribution network area load, vehicle arrival frequency, and rated power of each socket; and acquiring the charging area temperature, transient voltage of each socket, feeder impedance, and spatial coordinates includes: acquiring the distribution network area load and rated power of each socket through an area edge computing gateway; collecting vehicle arrival events through a visual sensing terminal deployed at the entrance of the charging area, and accumulating and calculating the vehicle arrival frequency within a preset time window; collecting the charging area temperature through an environmental sensing node array; at the initial conduction moment when the electric bicycle physically connects to each charging socket, using the high-frequency sampling circuit built into the socket to capture the voltage drop and recovery waveform within a preset period as the transient voltage of each socket; extracting spatial coordinates based on a pre-constructed three-dimensional grid topology of the charging station, and quantitatively calculating the feeder impedance based on the physical wire diameter and wiring physical length from each socket to the secondary side of the area transformer.

3. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The process of calculating the global capacity limit includes: obtaining the distribution network transformer capacity and expected fluctuating load; calculating the difference between the transformer capacity and the sum of the substation load and the expected fluctuating load to generate a dynamic power supply margin; multiplying the inrush frequency, preset time window and rated power to predict the tidal load increment; and comparing the dynamic power supply margin with the tidal load increment to determine the global capacity limit.

4. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The process of dividing the area into dual-feature area, single-feature area, and unconstrained area includes: extracting the full-load operating current of each socket under rated power; multiplying the full-load operating current with the corresponding feeder impedance to quantify and generate the theoretical maximum line loss voltage drop of each socket, which is used as the line voltage drop feature; using spatial coordinates to perform illumination tracing and shadow projection calculations in a preset three-dimensional building model of the station to extract the cumulative exposure time of each socket under direct sunlight, which is used as the sunlight exposure feature; and combining the line voltage drop feature with the sunlight exposure feature. A two-dimensional feature space is constructed, and extreme values ​​for voltage drop risk and exposure duration are set in the two-dimensional feature space. The set of sockets whose line voltage drop characteristics are greater than the extreme value for voltage drop risk and whose sunlight exposure characteristics are greater than the extreme value for exposure duration are divided into dual feature regions. The set of sockets whose only line voltage drop characteristics are greater than the extreme value for voltage drop risk or whose only sunlight exposure characteristics are greater than the extreme value for exposure duration are divided into single feature regions. The set of sockets whose line voltage drop characteristics are not greater than the extreme value for voltage drop risk and whose sunlight exposure characteristics are not greater than the extreme value for exposure duration are divided into unconstrained regions.

5. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The process of generating updated limits includes: traversing all sockets in each partition and marking the socket with the highest polarization scalar value as the reference socket; calculating the Euclidean distance between each socket and the reference socket based on the spatial coordinates; filtering out sockets whose Euclidean distance is less than the preset distance threshold to construct the adjacent set; dividing the Euclidean distance corresponding to each socket in the adjacent set by the preset distance threshold to generate a spatial thermal attenuation coefficient less than 1; and multiplying the original environmental limit of each socket in the adjacent set by the corresponding spatial thermal attenuation coefficient to calculate and generate the derated updated limit.

6. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The process of splitting the global capacity limit into power quotas corresponding to each zone includes: accumulating the updated limits and environmental limits of all sockets within the target charging area to generate a sum of limits for each socket; when the sum of limits for each socket is greater than the global capacity limit, for the dual-feature region, applying a first-level thermal impedance attenuation and voltage drop compensation pre-deduction to the global capacity limit, extracting the corresponding capacity to generate a first stringent quota; for the single-feature region, extracting the capacity corresponding to the current region and implementing a unilateral derating constraint to generate a second restricted quota; for the unconstrained region, subtracting the first stringent quota and the second restricted quota from the global capacity limit to obtain the remaining capacity limit, assigning it to the current region as a third regular quota.

7. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The evaluation model includes: a topology reconstruction layer, which maps and reconstructs all charging sockets into a virtual power supply topology tree with hierarchical parent-child relationships based on the branch connection relationships in the construction wiring topology data and the feeder impedance of each charging socket, and uses the polarization scalar as the attribute feature of the corresponding node; a line environment interaction layer, which classifies each target socket in the virtual power supply topology tree as a core node, and classifies the remaining charging sockets that share the same upstream power supply path as homogeneous environment nodes; counts the number of nodes in the homogeneous environment nodes whose polarization scalar is greater than that of the core node, calculates the ratio of the number of such nodes to the total number of homogeneous environment nodes, and generates a path congestion feature value; and a collaborative decision layer, which concatenates the polarization scalar of the target socket, its depth level in the virtual power supply topology tree, and the path congestion feature value into a context feature vector, inputs it into the classification network structure, and outputs the corresponding scheduling feature value; wherein, when the depth level is greater than a preset depth level and the path congestion feature value is greater than a preset congestion extreme value, the classification network structure is forced to output the scheduling feature value ranked last.

8. The electric bicycle charging scheduling method based on load prediction according to claim 1, characterized in that, The process of controlling the contactor to disconnect when the degradation characteristic value is greater than the preset degradation extreme value includes: mapping the scheduling characteristic value to a preset discrete power level table in descending order of value, querying and generating the control data corresponding to each socket; outputting a drive level based on the control data, controlling the working circuit relay array built into the socket to close to the physical conduction path of the corresponding level; after the physical conduction path is closed, starting a timer and continuing for a preset stable delay period, reading the voltage difference between the input and output terminals of the working circuit relay array at the end of the preset stable delay period as the steady-state voltage drop, and simultaneously collecting the actual working current on the physical conduction path; calculating the current contact resistance by dividing the steady-state voltage drop by the actual working current, calculating the absolute difference between the current contact resistance and the preset nominal contact resistance, and using this absolute difference as the degradation characteristic value characterizing the internal resistance state of mechanical contact erosion; when the degradation characteristic value is greater than the preset degradation extreme value, generating a hardware-level physical isolation instruction, and using the hardware-level physical isolation instruction to directly drive the main circuit contactor connected in series with the working circuit relay array to perform a mechanical disconnection action.

Images