A vehicle-to-pile interaction robust optimization scheduling method considering behavioral uncertainty
By constructing a multi-dimensional feature vector index library and a multi-objective optimization scheduling model, the problem of accurately representing discrete load characteristics in the electric vehicle-charging pile interaction scenario was solved, and efficient, stable operation and real-time control of electric vehicle charging facilities were realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- STATE GRID GANSU ELECTRIC POWER CO JIUQUAN POWER SUPPLY CO
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-19
Smart Images
Figure CN121880692B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a robust optimization scheduling method for vehicle-charging station interaction that takes into account behavioral uncertainties, belonging to the field of electric vehicle energy management and charging control technology. Background Technology
[0002] In current scenarios where electric vehicles are massively integrated into charging stations, random fluctuations in vehicle-charging pile loads impact the safe and stable operation of the charging station. To address this issue, robust optimization-based scheduling strategies have become the mainstream approach. Typically, an uncertainty set is constructed to describe the range of user behavior fluctuations. A probability density function derived from historical operational data is selected, and combined with a preset confidence level, a mathematical set with geometric shapes such as a box or ellipsoid is determined to cover potential load deviations. The charging management system then formulates charging and discharging power commands based on the boundaries of this set, ensuring that within the fluctuation range, the vehicle-charging pile interaction system, while satisfying the vehicle battery state of charge constraints, ensures that the charging pile output power meets the preset design parameters. While implementing boundary constraints, existing robust optimization scheduling strategies often suffer from a disconnect between mathematical modeling and physical reality when attempting to address the aforementioned uncertainties. Given fixed hardware facilities, the accuracy of the control algorithm's characterization of the safety boundary directly determines the carrying capacity of the station area. However, existing technologies still have mechanistic limitations when constructing uncertainty sets. For example, Chinese invention patent CN114944662B discloses a robust optimization scheduling method for electric vehicle cluster grid connection based on support vector clustering. This method optimizes the shape of the uncertainty set through machine learning algorithms and uses support vector clustering to construct a minimum hypersphere containing sample data, aiming to replace the traditional box set.
[0003] However, when the aforementioned modeling method based on the assumption of continuous probability is applied to high-density vehicle-charging station interaction scenarios, its inherent mechanistic limitations are exposed. The access, disconnection, and charging demands of electric vehicle users are physically discrete and impulsive. Collective access or random disconnection within a specific time period generates power jumps rather than a smooth probability distribution. In order to mathematically encompass these nonlinear jump points, the geometric uncertainty set in existing technologies usually adopts the method of expanding the set volume. Although this approach ensures stability, it leads to excessive expansion of the safety boundary in engineering practice, causing the scheduling model to estimate uncertainty beyond the actual fluctuation range of the physical system. In this environment, there are inherent contradictions in the existing technology system that are difficult to reconcile. If the regular geometric set is used to cover the discrete impulsive characteristics, the scheduling strategy becomes conservative, and the effective capacity of key equipment such as transformers is occupied by redundant safety margins, resulting in a decrease in the utilization rate of power assets. If a high-order mixed probability model is introduced to approximate the real discrete distribution, the variable dimension of the optimization problem explodes, and its solution time cannot meet the real-time control requirements of the station area at the minute level.
[0004] Therefore, the technical problem to be solved by this invention is how to establish a scheduling method that can accurately characterize discrete interactive behavior features and precisely define the safety boundary of station operation while meeting timeliness requirements. Summary of the Invention
[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty, comprising the following steps:
[0006] Step 101: Obtain historical operating condition data of the electric vehicle. The historical operating condition data includes the connection time, disconnection time, and initial state of charge.
[0007] Step 102: Extract multi-dimensional feature vectors from historical operating condition data and construct an operating condition index library;
[0008] Step 103: Obtain the current operating condition feature vector of the electric vehicle to be scheduled in real time, and use the hash matching algorithm to retrieve the K sets of historical operating condition trajectory points with the smallest Hamming distance to the current operating condition feature vector in the operating condition index library.
[0009] Step 104, Extract The maximum and minimum power values of historical operating condition trajectory points at each scheduling time are used to construct the power fluctuation envelope interval;
[0010] Step 105: Establish a multi-objective optimization scheduling model with the objectives of maximizing the stability of the total charging load curve in the station area and minimizing the charging cost on the user side. Use the power fluctuation envelope interval as the boundary of the feasible region for optimization calculation in the multi-objective optimization scheduling model, and generate vehicle-pile cooperative charging control commands.
[0011] Preferably, step 102 specifically includes: using multiple sets of random linear projection functions to perform dimensionality reduction processing on historical operating condition data to generate hash codes corresponding to each historical operating condition data; classifying data points with the same hash code into the same hash bucket to establish an operating condition index library; in step 103, when retrieving K sets of historical operating condition trajectory points, calculating the hash code of the current operating condition feature vector to locate the target hash bucket, and performing operating condition similarity comparison within the target hash bucket.
[0012] Preferably, when establishing the operating condition index library in step 102, the operating condition mapping table is also constructed; the operating condition mapping table records the index relationship between each hash bucket and the physical trajectory data in the historical operating condition data; the physical trajectory data includes the historical charging and discharging power time series corresponding to each historical operating condition data.
[0013] Preferably, the method for constructing the power fluctuation envelope interval in step 104 is as follows: at each scheduling time... Next, calculate The maximum value in the set of charge and discharge power corresponding to the historical operating condition trajectory points. and minimum value The power fluctuation envelope interval U(t) satisfies the following judgment rule: Where P(t) is the power value of the electric vehicle to be scheduled at scheduling time t. For the historical working condition trajectory points of group K at time... Maximum power value, Let t be the minimum power value of the historical operating condition trajectory points of group K at time t.
[0014] Preferably, in step 105, a rolling time-domain optimization strategy is used to solve the multi-objective optimization scheduling model. The rolling time-domain optimization strategy includes: updating the current operating condition feature vector in real time as the scheduling time progresses, and repeating steps 103 and 104 to achieve online correction of the power fluctuation envelope interval.
[0015] Preferably, the multi-objective optimization scheduling model also includes a charge state continuity constraint; the charge state continuity constraint is established based on the real-time charge state of the electric vehicle to be scheduled, the power fluctuation envelope range, and the battery charging and discharging efficiency, and is limited to the electric vehicle to be scheduled reaching a preset target charge value at the time of off-grid.
[0016] Preferably, the multi-objective optimization scheduling model sets weighting factors based on the load characteristics of vehicle-to-pile interaction; the load characteristics are determined based on the combination ratio of fast charging, slow charging, and vehicle discharge to the station area; and the weighting factors are used to adjust the optimization priority between the station area load smoothing control objective and the user-side cost objective.
[0017] Preferably, the vehicle-charging pile interaction scheduling instruction generated in step 105 includes the active power adjustment amount at each scheduling time; the active power adjustment amount is used to control the power output of the controlled charging pile to offset the power deviation caused by the fluctuation of the basic power load in the station area.
[0018] Preferably, the method further includes a data incremental update step: the current operating condition feature vector and its corresponding measured charging and discharging power trajectory after executing the vehicle-pile interaction scheduling instruction are incorporated into the operating condition index library as a reference for executing step 104 in subsequent scheduling cycles.
[0019] Preferably, in step 105, when solving the multi-objective optimization scheduling model, the power fluctuation envelope interval is mapped to the solution space boundary of the solver, restricting the scheduling instruction vector to be generated to fall within the power fluctuation envelope interval at each scheduling time.
[0020] Compared with the prior art, the beneficial effects of the present invention are:
[0021] 1. In the robust optimization scheduling method for vehicle-charging pile interaction, the problem of distortion in the representation of discrete pulse loads by traditional probabilistic models is solved, improving the charging and discharging control accuracy of the vehicle-charging pile interaction system and the operational stability of charging facilities under discrete load impacts. Existing technologies usually assume that the access behavior of electric vehicles follows a continuous probability distribution, which makes it impossible to accurately describe the characteristics of discrete pulse loads with suddenness and clustering. This invention adopts a working condition mapping mechanism based on multi-dimensional feature vectors to extract real load trajectories from the historical operating database that are of the same origin as the current spatiotemporal state. By superimposing the physical extrema of these discrete trajectories in the time domain, a non-convex dynamic safety envelope domain is constructed. This eliminates the mathematical smoothing of intermediate states, enabling the generated constraint boundary to accurately cover transient power jumps caused by collective user access or specific events. The power commands formulated by the charging management system accordingly can ensure that the output power of the vehicle-charging pile interaction system is always limited within the physical safety threshold of the charging facilities and vehicle batteries when facing nonlinear load impacts, effectively avoiding the risk of equipment exceeding the limit due to model assumption deviations.
[0022] 2. This invention overcomes the inherent contradiction between safety margin and equipment utilization in traditional robust optimization, improving the power utilization efficiency of existing charging facilities at charging stations and the economic operation level of charging services. In the construction of traditional box-type or ellipsoidal uncertainty sets, in order to cover low-probability extreme scenarios, it is often necessary to reserve excessive safety margins, which forces the actual load rate of key equipment such as transformers to be reduced, resulting in idle assets. This invention locks a limited set of historical scenarios that are highly related to the current environment through topological matching of high-dimensional features, thereby converging the boundary of uncertainty constraints from a broad statistical space to a compact physical feasible region. This data-driven boundary contraction mechanism reduces the ineffective reserve capacity space without reducing the system safety level, enabling charging stations to accept more vehicle charging needs and improving the utilization efficiency and economic operation level of the power supply and distribution facilities within the station.
[0023] 3. A lightweight edge computing architecture is established to meet the timeliness requirements of real-time control of large-scale vehicle-charging pile clusters. Addressing the problem of exponential growth in centralized optimization computation caused by large-scale vehicle-charging pile access, this invention transforms the complex real-time uncertainty quantification process into a low-latency feature library indexing and state matching process through feature discretization and state identifier mapping technology. This avoids the time-consuming probability density function integration and large-sample Monte Carlo simulation in traditional methods, reducing the computational load on the charging management controller. Combined with a hierarchical solution strategy for rolling time-domain optimization, the system can dynamically correct the power command of the entire network at a time granularity of seconds, ensuring that the scheduling strategy can track photovoltaic power output fluctuations and changes in user random behavior in real time, thus guaranteeing power balance and energy supply-demand matching at the charging station level. Attached Figure Description
[0024] Figure 1 A flowchart illustrating the specific implementation of the robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainties in this invention.
[0025] Figure 2 This is a schematic diagram of the overall architecture and data interaction closed loop of the vehicle-charging pile interaction robust optimization charging management system of the present invention.
[0026] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0027] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.
[0028] A robust optimization scheduling method for vehicle-to-pile interaction considering behavioral uncertainty includes the following steps:
[0029] Step 101: Obtain historical operating condition data of the electric vehicle. The historical operating condition data includes the connection time, disconnection time, and initial state of charge.
[0030] Step 102: Extract multi-dimensional feature vectors from historical operating condition data and construct an operating condition index library;
[0031] Step 103: Obtain the current operating condition feature vector of the electric vehicle to be scheduled in real time, and use the hash matching algorithm to retrieve the K sets of historical operating condition trajectory points with the smallest Hamming distance to the current operating condition feature vector in the operating condition index library.
[0032] Step 104, Extract The maximum and minimum power values of historical operating condition trajectory points at each scheduling time are used to construct the power fluctuation envelope interval;
[0033] Step 105: Establish a multi-objective optimization scheduling model with the objectives of maximizing the stability of the total charging load curve in the station area and minimizing the charging cost on the user side. Use the power fluctuation envelope interval as the boundary of the feasible region for optimization calculation in the multi-objective optimization scheduling model, and generate vehicle-pile cooperative charging control commands.
[0034] Preferably, step 102 specifically includes: using multiple sets of random linear projection functions to perform dimensionality reduction processing on historical operating condition data to generate hash codes corresponding to each historical operating condition data; classifying data points with the same hash code into the same hash bucket to establish an operating condition index library; in step 103, when retrieving K sets of historical operating condition trajectory points, calculating the hash code of the current operating condition feature vector to locate the target hash bucket, and performing operating condition similarity comparison within the target hash bucket.
[0035] Preferably, when establishing the operating condition index library in step 102, the operating condition mapping table is also constructed; the operating condition mapping table records the index relationship between each hash bucket and the physical trajectory data in the historical operating condition data; the physical trajectory data includes the historical charging and discharging power time series corresponding to each historical operating condition data.
[0036] Preferably, the method for constructing the power fluctuation envelope interval in step 104 is as follows: at each scheduling time t, calculate the maximum value in the charging and discharging power set corresponding to the K groups of historical operating condition trajectory points. and minimum value The power fluctuation envelope interval U(t) satisfies the following judgment rule: Where P(t) is the power value of the electric vehicle to be scheduled at scheduling time t. For the historical working condition trajectory points of group K at time... Maximum power value, Let t be the minimum power value of the historical operating condition trajectory points of group K at time t.
[0037] Preferably, in step 105, a rolling time-domain optimization strategy is used to solve the multi-objective optimization scheduling model. The rolling time-domain optimization strategy includes: updating the current operating condition feature vector in real time as the scheduling time progresses, and repeating steps 103 and 104 to achieve online correction of the power fluctuation envelope interval.
[0038] Preferably, the multi-objective optimization scheduling model also includes a charge state continuity constraint; the charge state continuity constraint is established based on the real-time charge state of the electric vehicle to be scheduled, the power fluctuation envelope range, and the battery charging and discharging efficiency, and is limited to the electric vehicle to be scheduled reaching a preset target charge value at the time of off-grid.
[0039] Preferably, the multi-objective optimization scheduling model sets weighting factors based on the load characteristics of vehicle-to-pile interaction; the load characteristics are determined based on the combination ratio of fast charging, slow charging, and vehicle discharge to the station area; and the weighting factors are used to adjust the optimization priority between the station area load smoothing control objective and the user-side cost objective.
[0040] Preferably, the vehicle-charging pile interaction scheduling instruction generated in step 105 includes the active power adjustment amount at each scheduling time; the active power adjustment amount is used to control the power output of the controlled charging pile to offset the power deviation caused by uncontrolled load fluctuations in the station area.
[0041] Preferably, the method further includes a data incremental update step: the current operating condition feature vector and its corresponding measured charging and discharging power trajectory after executing the vehicle-pile interaction scheduling instruction are incorporated into the operating condition index library as a reference for executing step 104 in subsequent scheduling cycles.
[0042] Preferably, in step 105, when solving the multi-objective optimization scheduling model, the power fluctuation envelope interval is mapped to the solution space boundary of the solver, restricting the scheduling instruction vector to be generated to fall within the power fluctuation envelope interval at each scheduling time.
[0043] Example 1: This example is applied to a centralized charging station for electric vehicles in an underground parking lot of an urban commercial complex. The main power supply circuit terminal of the station in this area is in a critical state of heavy load, and the user charging behavior exhibits multi-peak discrete jump characteristics for specific time points between 17:00 and 19:00. Under this condition, the robust uncertainty set generated by the stochastic programming method based on the assumption of continuous probability distribution covers the probability interval of the intermediate state that does not exist physically, causing the transformer capacity to be occupied by redundant safety margins, which in turn causes vehicles waiting to be charged to be rejected due to insufficient nominal capacity. The charging management system of this invention initiates a discrete scene retrieval process based on feature vector inverted index. The data acquisition terminal reads the operating status data of the electric vehicles currently connected and waiting to be scheduled. The feature extraction unit converts the operating status data into a set of multi-dimensional feature vectors, including the access time slot index, initial state of charge, required power and current time-of-use electricity rate code. The processor executes the locality-sensitive hash algorithm and uses multiple sets of preset random linear projection functions to perform dimensionality reduction projection and binarization processing on the feature vector to generate the corresponding binary hash code, which is used as the key value of the inverted index library and located in the corresponding hash bucket in the memory.
[0044] Based on a pre-established index relationship, the system extracts K sets of historical real charge and discharge power time series sequences from the hash bucket that have a Hamming distance less than a preset threshold. In this embodiment, K is set to 50. The system physically overlays these 50 sets of historical trajectories in a time-domain coordinate system and extracts the power maxima among all trajectory points at each scheduling time t. With power minimum ,in, This represents the historical maximum power at time t. The two extreme boundaries, representing the historical minimum power at time t, form a non-convex power fluctuation envelope interval U(t) in the time domain. By directly stacking physical data, intermediate states that exist in the statistical space but have not occurred in physical reality are eliminated. The feasible domain boundary of the uncertainty constraint is converged from the convex set to the edge of the physical behavior. A multi-objective optimization scheduling model is established with the goal of optimizing the stability of the total charging load curve of the station area and minimizing the charging cost on the user side. The power fluctuation envelope interval U(t) is mapped to the linear inequality constraint set of the solver. Since the constraint boundary excludes the invalid probability space, the effective carrying capacity of the station area is numerically released. Under the premise of satisfying the safety constraints of the station bus voltage and facility current carrying capacity, the solver calculates and generates the vehicle-pile interaction scheduling instruction. This instruction contains the active power adjustment amount, which is used to control the power output of the controlled charging pile. After the instruction is executed, the system uses the measured feature vector and charging and discharging power trajectory of this scheduling as data increments, and after hash encoding, it is incorporated into the working condition index library.
[0045] Example 2: This example constructs a simulation experimental platform containing 120 AC slow charging piles (rated power 7kW) and 20 DC fast charging piles (rated power 60kW) to verify the effectiveness of the robust optimization scheduling method proposed in this invention in dealing with highly uncertain user charging behavior. The platform is built based on an internal power distribution topology model containing multiple charging units, with node 18 as the charging station access point. To simulate random disturbances in real industrial scenarios, a Gaussian distribution load prediction error with a standard deviation of 15% of the predicted value and a 2% random communication packet loss rate are introduced in the data generation stage. The experimental data comes from the anonymized operation logs of a municipal-level charging operation management platform in 2023, with a sampling interval of 15 minutes, covering 30 consecutive working days. The key parameter K (i.e., the number of historical operating condition trajectory points selected) in the locality-sensitive hash retrieval is optimized and calibrated. The value of parameter K directly affects the stability and economy of the system. Too small a value of K will result in insufficient retrieval samples, an overly narrow power fluctuation envelope, and reduced stability. On the other hand, too large a value of K will introduce irrelevant noise samples, leading to an overly conservative envelope and reduced economy. Therefore, the value of K is set to be between 10 and 100, with a step size of 10. 100 Monte Carlo simulations are run at each value level. The experimental results show that when the value of K increases from 10 to 50, the probability of node voltage exceeding the limit drops sharply from 8.5% to 0.2%. When it continues to increase to 100, the probability of exceeding the limit only decreases slightly to 0.15%, but the operating cost increases by 12.4%. Based on this performance inflection point, this embodiment determines the value of K to be 50.
[0046] To verify the technical effectiveness of this invention, three sets of comparative experiments were set up: Control group 1 adopted a deterministic optimization scheduling strategy based on the day-ahead prediction curve; Control group 2 adopted a traditional robust optimization method based on the construction of a box uncertainty set based on statistical variance; The sample group of this invention adopted a dynamic power fluctuation envelope interval generated based on inverted index and local sensitive hashing for scheduling. The experiment simulated the evening peak period from 17:00 to 19:00. During this period, because Control group 1 relied only on a single prediction curve, when it encountered a pulse impact where the actual load was 35% higher than the predicted value (such as at 17:45), the voltage of node 18 dropped to 0.88pu, which was lower than the safety threshold of 0.90pu, resulting in the forced disconnection of some loads. Although Control group 2 introduced robust constraints, the regular hyperrectangular set it constructed contained a large number of physically non-existent extreme corner states, which severely compressed the system's schedulable domain. The average charging power was only 45% of the rated power, and the user satisfaction score was as low as 6.2 points (out of 10).
[0047] In contrast, the sample group of this invention, by retrieving the 50 historical trajectories with the smallest Hamming distance to the current feature vector in real time, constructs a power fluctuation envelope interval U(t) that accurately matches the non-convex distribution characteristics of the actual load. During the load impact at 17:45, the upper boundary of the envelope interval... The pulse was covered, the scheduling command responded in advance, and the node voltage was stabilized above 0.93 pu. At the same time, due to the elimination of invalid probability space, the effective carrying capacity of the system was improved by 28% compared with the control group 2, the average charging power reached 73% of the rated power, and the user satisfaction score improved to 9.1 points. In addition, key intermediate data showed that the invention used the inverted index mechanism to reduce the average retrieval time of a single scheduling request from 1250ms of traditional full database traversal to 18ms, achieving millisecond-level real-time response and supporting the online construction of dynamic envelope domains. The above data strongly proves that the invention improves the system's operating efficiency and user experience while ensuring the security of the site domain.
[0048] Example 3: This example supplements the key aspects of the Locality Sensitive Hash (LSH) retrieval and multi-objective optimization scheduling model with specific engineering implementation procedures and parameter adaptive logic, eliminating the potential technical black box of how to determine the optimal number of projection functions and how to balance stability and economic weights. Before the dimensionality reduction process in step 102, Z-Score-based dimensionless standardization is performed on the multi-dimensional feature vectors of historical operating condition data to eliminate the retrieval weight bias caused by differences in physical magnitudes. The processor traverses the access time index, initial state of charge, and demand power data of the samples in the database and calculates the global arithmetic mean of the data in each dimension. with standard deviation According to the formula Mapping the original numerical value x of the feature vector transforms heterogeneous data from different physical units into standard normal distribution values with a mean of 0 and a variance of 1. The process generates standardized vectors as input to a random linear projection function, ensuring that the Hamming distance calculation depends on the topological similarity of the waveform under the operating conditions, rather than being controlled by the absolute amplitude of high numerical dimensions. A random linear projection function is constructed, and the initial random matrix is decoupled using the Schmitt orthogonalization algorithm to guarantee the independence of hash code bits. The initialization program generates a random matrix whose elements follow a standard normal distribution N(0,1). The matrix is traversed column by column, and the projection components of previously processed column vectors are subtracted from the column vectors. The resulting vectors are normalized to obtain pairwise orthogonal unit basis vectors. These orthogonal basis vectors are established as linear projection coefficients, dividing the high-dimensional feature space into non-overlapping sub-regions. This avoids the abnormally high collision rate of hash buckets caused by the linear correlation of randomly generated vectors, enabling the maximum spatial discriminability within the limited encoding length of the index library.
[0049] Regarding the determination of the number of projection functions L in LSH retrieval, this invention abandons the traditional approach of setting them solely based on experience, and introduces a parameter adaptive calibration procedure based on the dual constraints of collision probability and retrieval efficiency, defining a collision probability threshold that characterizes retrieval accuracy. Its physical meaning is: for any two feature vectors whose distance in Euclidean space is less than a preset threshold R, the probability that they are mapped to the same hash code is no less than... This probability exhibits a monotonically non-linear relationship with the number of projection functions L. Secondly, an upper limit for the hash bucket capacity, representing the computational load, is defined. Its physical meaning is: to ensure millisecond-level retrieval response, the number of historical trajectory indexes stored in any hash bucket should not exceed [a certain limit]. During the system initialization phase, the following offline calibration process is executed: At least 10,000 sample vectors are randomly selected from the historical database to construct a test set. The initial value of L is set to 1, and incremented by 1 step. At each L value, the actual collision probability of sample pairs in the test set that satisfy the distance threshold R is calculated, and the average capacity of each hash bucket is calculated. When the calculated actual collision probability is first greater than or equal to... And the capacity of all hash buckets is no higher than At that time, the current L value is locked as the optimal projection dimension for the application scenario. Through this procedure, the abstract parameter setting is transformed into a closed, data-feedback-based deterministic optimization process.
[0050] To address the weight allocation problem between station area objectives (power fluctuation) and user-side objectives (charging cost) in multi-objective optimization scheduling models, this invention constructs a dynamic weight adjustment mechanism based on the curvature of the Pareto front. Traditionally, fixed weighting coefficients are used, which are difficult to adapt to the dynamic changes in station area safety and user demand at different times. In this embodiment, the system quickly generates a set of non-dominated solutions based on the current station area load status, such as the load rate of the station's main power supply line and the real-time electricity price. The curvature of the Pareto front curve formed by this non-dominated solution set in the objective function space is calculated. The point with the largest curvature, known as the knee, mathematically represents the critical state where the marginal rate of substitution between the two objectives changes drastically. After this point, a small improvement in one objective requires a significant deterioration in the other. The system uses the weight combination corresponding to this knee as the optimal weight vector for the current scheduling cycle. When the station area is under heavy load, the shape of the Pareto front will change. The system exhibits a dynamic distortion, shifting the knee point towards a direction prioritizing station safety. Conversely, when the station is lightly loaded, the knee point shifts towards a direction prioritizing user costs. By tracking this mathematical characteristic in real time, the system can adaptively find the optimal balance between ensuring safety and reducing costs without manual intervention. Furthermore, to ensure the physical executability of generated scheduling instructions, a dead-zone logic for equipment actions is explicitly embedded in the constraints of the optimization model. Considering that frequent adjustments to charging pile power may shorten relay lifespan or cause overheating of electronic components, a minimum action interval constraint is introduced into the model. If the change in power adjustment calculated at the current moment is less than the preset dead-zone threshold (e.g., 5% of the rated power), the instruction from the previous moment remains unchanged. This is achieved by introducing 0-1 variables based on state preservation, avoiding high-frequency invalid actions of equipment caused by minute numerical fluctuations, thus enhancing the practicality and durability of the solution in engineering sites.
[0051] Example 4: This example addresses the initial state uncertainties and equipment aging drift risks that may be encountered when deploying the solution in different industrial sites. It establishes a standardized pre-deployment calibration and adaptive reconfiguration procedure, making it a mandatory pre-operation step before the system is officially put into operation. This eliminates the risk of model failure due to environmental differences and ensures the stability and consistency of the technical solution throughout its entire lifecycle. Regarding the cold start issue of newly connected charging stations, the system executes a baseline load profile construction process based on typical days. During the first 7 calendar days after the system's initial power-on, the scheduling module is in passive monitoring mode, only collecting data without interruption. Control commands are issued, and during this period, the data acquisition terminal records the full load data of the site at a high frequency, such as once per minute, including the start and stop times of each charging pile, power curves, and user dwell time. The processor uses the sliding window statistical method to calculate the load baseline mean and variance of the site in each time period, and generates an initial operating condition feature vector library accordingly. When the number of accumulated samples reaches a preset threshold, such as 1000 valid records, the system automatically triggers the parameter initialization program of the local sensitive hash function. Based on the distribution characteristics of the actual data, the initial direction of the projection vector is determined using the principal component analysis method, thereby completing the cold start construction of the inverted index library.
[0052] To address parameter drift caused by equipment aging during long-term operation, the system incorporates a periodic closed-loop feedback calibration mechanism. The calibration cycle is set to 30 calendar days. At the end of each calibration cycle, the system automatically extracts all actual scheduling commands and corresponding station response data, such as the improvement in the output voltage stability of power supply facilities. If the control error (i.e., the deviation between the expected voltage value and the measured voltage value) exceeds the preset safety tolerance of ±2% for three consecutive scheduling cycles, it is determined that the system model parameters have drifted. At this time, the system automatically starts the online correction program, uses the measured data of the most recent cycle to fine-tune and update the network loss coefficient matrix in the multi-objective optimization model, and recalculates the knee position of the Pareto front to dynamically adapt to the current aging state of the equipment. Through this closed-loop calibration process, it is ensured that the scheduling accuracy under continuous operation conditions does not decay over time.
[0053] Example 5: This example addresses the historical operating condition data collection process in the solution by establishing a standardized data quality cleaning and outlier correction procedure. This resolves the issue of unusable raw data due to equipment failure or communication interference, ensuring the reliability of the inverted index database construction. For potential missing values, outliers, and logical conflicts in the historical operating condition data, the system performs a three-level preprocessing process before data entry. The first level is physical constraint verification, where the system, based on the battery management system (BMS) hardware parameters, removes all records that violate physical limits, such as SOC values less than 0% or greater than 100%, or charging currents exceeding 1.2 times the rated value. The second level is time sequence continuity repair. For short-term data loss (less than 15 minutes) caused by momentary communication interruptions, the system uses Lagrange interpolation to complete the data based on adjacent valid data points. For long-term data loss (more than 15 minutes), the complete record of the charging event is directly removed to avoid introducing significant artificial synthesis errors. The third level is feature consistency cleaning, where the system performs logical correlation checks on the voltage, current, and SOC change rate in the same charging event, removing abnormal samples that violate electrochemical laws, such as constant voltage but drastic SOC jumps. Through this procedure, the validity of the original dataset is increased from 85% to over 98%, providing a high-quality data foundation for subsequent feature extraction and hash mapping.
[0054] This embodiment clarifies the specific quantification formula and parameter value sources for calculating user-side charging costs in a multi-objective optimization scheduling model. Defined as the sum of electricity cost and time penalty cost, where electricity cost is calculated based on local time-of-use pricing policies, dividing the day into peak, off-peak, and valley periods, each corresponding to different electricity rates; and time penalty cost is used to quantify the psychological loss incurred by users due to charging delays, and its calculation formula is as follows: ,in The actual time to complete charging. The expected disconnection time for users is α, and the time sensitivity coefficient is α. In this embodiment, questionnaires and behavioral experiments were conducted on different types of users, such as ride-hailing drivers and private car owners, to determine the empirical range of α. In actual scheduling, the corresponding coefficient value was automatically matched according to the user profile tags, eliminating the ambiguity of user satisfaction quantification in the model and ensuring the objectivity and interpretability of the multi-objective optimization results.
[0055] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty, characterized in that, Includes the following steps: Step 101: Obtain historical operating condition data of the electric vehicle. The historical operating condition data includes the connection time, disconnection time, and initial state of charge. Step 102: Extract multidimensional feature vectors from historical operating condition data and construct an operating condition index library; use multiple sets of random linear projection functions to perform dimensionality reduction processing on historical operating condition data to generate hash codes corresponding to each historical operating condition data; classify data points with the same hash code into the same hash bucket to establish an operating condition index library; also includes constructing an operating condition mapping table; the operating condition mapping table records the index relationship between each hash bucket and the physical trajectory data in the historical operating condition data; Physical trajectory data includes historical charging and discharging power time series corresponding to each historical operating condition; Step 103: Obtain the current operating condition feature vector of the electric vehicle to be scheduled in real time, and use a hash matching algorithm to search the operating condition index database for the element with the smallest Hamming distance to the current operating condition feature vector. Collect historical work condition trajectory points; calculate the hash code of the current work condition feature vector to locate the target hash bucket, and perform work condition similarity comparison within the target hash bucket; Step 104, Extract The power maximum and minimum values of historical operating condition trajectory points at each scheduling time are used to construct the power fluctuation envelope interval. The method for constructing the power fluctuation envelope interval is as follows: at each scheduling time... Next, calculate the maximum value in the set of charging and discharging power corresponding to the K groups of historical operating condition trajectory points. and minimum value The power fluctuation envelope interval U(t) satisfies the following judgment rule: Where P(t) represents the electric vehicle to be scheduled at the scheduling time. The execution power value, For the historical working condition trajectory points of group K at time... Maximum power value, Let K be the minimum power value of the historical working condition trajectory points at time t; Step 105: Establish a multi-objective optimization scheduling model with the objectives of maximizing the stability of the total charging load curve in the station area and minimizing the charging cost on the user side. Use the power fluctuation envelope interval as the boundary of the feasible region for optimization calculation of the multi-objective optimization scheduling model to generate vehicle-pile coordinated charging control commands. When solving the multi-objective optimization scheduling model, the power fluctuation envelope interval is mapped as the solution space boundary of the solver, restricting the scheduling command vector to be generated to fall within the power fluctuation envelope interval at each scheduling time.
2. The robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty according to claim 1, characterized in that, In step 105, a rolling time-domain optimization strategy is used to solve the multi-objective optimization scheduling model. The rolling time-domain optimization strategy includes: updating the current operating condition feature vector in real time as the scheduling time progresses, and repeating steps 103 and 104 to achieve online correction of the power fluctuation envelope interval.
3. The robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty according to claim 1, characterized in that, The multi-objective optimization scheduling model also includes a charge state continuity constraint. The charge state continuity constraint is established based on the real-time charge state of the electric vehicle to be scheduled, the power fluctuation envelope range, and the battery charging and discharging efficiency, and is limited to the electric vehicle to be scheduled reaching the preset target charge value at the time of off-grid.
4. The robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty according to claim 1, characterized in that, The multi-objective optimization scheduling model sets weighting factors based on the load characteristics of vehicle-pile interaction; the load characteristics are determined based on the combination ratio of fast charging, slow charging and vehicle discharge to the station area; the weighting factors are used to adjust the optimization priority between the station area load smoothing control objective and the user-side cost objective.
5. A robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty as described in claim 1, characterized in that, The vehicle-charging pile interaction scheduling instruction generated in step 105 includes the active power adjustment amount at each scheduling time; the active power adjustment amount is used to control the power output of the controlled charging pile to offset the power deviation caused by the fluctuation of the basic power load in the station area.
6. The robust optimization scheduling method for vehicle-pile interaction considering behavioral uncertainty according to claim 1, characterized in that, The method also includes an incremental data update step: the current operating condition feature vector and its corresponding measured charging and discharging power trajectory after executing the vehicle-pile interaction scheduling instruction are incorporated into the operating condition index library as a reference for executing step 104 in subsequent scheduling cycles.