A method for demand forecasting and layout optimization of electric vehicle charging stations

Abstract

This invention discloses a method for demand forecasting and layout optimization of electric vehicle charging stations. It includes: dual-dimensional uncertainty quantification, which uses a demand forecasting model and a fault prediction model to predict the probability confidence intervals of charging demand fluctuations and equipment failures, respectively, and couples these to generate a joint uncertainty scenario set; robust redundancy layout solution, which constructs a multi-objective optimization function with the joint scenario set as constraints, and uses an improved genetic algorithm with dynamically allocated weights to solve the problem, obtaining configuration schemes for hot backup and cold backup redundant charging stations; and dynamic iterative optimization, which updates the model and scenario set based on real-time data and dynamically adjusts the layout scheme accordingly, forming a closed loop. This invention achieves an optimal balance between charging service continuity, resource utilization, and economy under the dual uncertainties of demand fluctuations and equipment failures.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging infrastructure management technology, and in particular to a method for predicting the demand and optimizing the layout of electric vehicle charging stations. Background Technology

[0002] With the continuous growth of electric vehicle ownership, the planning and layout of charging infrastructure has become a crucial foundation for supporting the large-scale application of electric vehicles. The number, spatial distribution, and operational reliability of charging piles directly affect user charging convenience, charging service continuity, and the overall utilization efficiency of related resources. At the same time, the construction of charging facilities needs to consider investment costs, operation and maintenance costs, and long-term operational benefits, thus placing higher demands on charging pile demand forecasting and layout optimization.

[0003] In real-world operating environments, the demand for electric vehicle charging exhibits significant spatiotemporal fluctuations. Charging demand in different regions and at different times is influenced by a combination of factors, resulting in instability and uncertainty. Furthermore, as long-term operating electrical equipment, the performance of charging piles changes with usage time, load levels, and environmental conditions. Equipment failures or performance degradation are inevitable, further increasing the uncertainty on the supply side of charging services.

[0004] However, existing methods for demand analysis and deployment planning of charging stations often fail to fully reflect these complex characteristics. On the one hand, demand analysis is largely based on deterministic results, making it difficult to effectively characterize the fluctuation range and risk level of demand during future operation. This can easily lead to a mismatch between service capacity and actual demand in practical applications. On the other hand, deployment planning often assumes that charging facilities can operate continuously and stably, lacking a systematic consideration of changes in equipment operating status and potential failures. This can result in insufficient reliability of charging services in some operating scenarios, or the adoption of over-configuration strategies to mitigate risks, leading to resource waste.

[0005] Furthermore, existing solutions generally lack the ability to comprehensively analyze the combined effects of multiple uncertainties. When demand fluctuations and equipment malfunctions occur simultaneously, the adaptability and robustness of existing layout schemes are clearly insufficient, making it difficult to guarantee the continuity of charging services and overall operational efficiency. In addition, the relevant planning results are mostly based on one-time static configurations, lacking a mechanism for continuous correction and dynamic adjustment based on operational data, making it difficult for the layout of charging facilities to adapt to changes in regional development and evolving operating conditions in the long term.

[0006] Therefore, there is an urgent need for a method that can comprehensively analyze changes in charging demand and uncertainties in equipment operation under complex operating environments, and on this basis, continuously optimize the layout of charging facilities to improve the service reliability, resource utilization efficiency and overall operational stability of the charging system. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a method for predicting the demand and optimizing the layout of electric vehicle charging stations, which can achieve a robust balance between charging service continuity, resource utilization and total life cycle cost in complex scenarios with multiple uncertainties.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for demand forecasting and layout optimization of electric vehicle charging stations, comprising: Step S1, Two-dimensional uncertainty quantification: Collect charging demand data and charging pile operation data of the target area, perform preprocessing and feature extraction to obtain demand feature set and equipment feature set; Based on the aforementioned demand feature set, the probability distribution and demand confidence interval of future charging demand fluctuations are predicted using a demand prediction model based on a Bayesian neural network. Based on the device feature set, the fault confidence interval of the charging pile's fault probability and fault duration in the future time period is predicted by the fault prediction model. Couple the demand confidence interval with the fault confidence interval to generate a set of joint uncertainty scenarios that includes multiple joint scenarios and their occurrence probabilities; Step S2, Solving for robust redundancy layout: Using the aforementioned set of joint uncertainty scenarios as constraints, a multi-objective optimization function is constructed with charging service satisfaction rate, resource utilization rate, and total life cycle cost as objectives. An improved non-dominated sorting genetic algorithm is used to solve the function. The algorithm dynamically allocates uncertainty weights based on the intensity of demand fluctuations and the degree of equipment aging. Based on the solution results, the configuration scheme of redundant charging piles in the target area is determined by the redundancy configuration model, including the type, quantity and layout location of hot backup redundant piles and cold backup redundant piles; Step S3, Dynamic Iterative Optimization: After the implementation of the plan, dynamic demand and equipment operation data of the target area will be collected in real time. Based on this dynamic data, the demand prediction model and the fault prediction model are updated on a rolling basis to update the demand confidence interval, the fault confidence interval, and the joint uncertainty scenario set; The configuration scheme of the redundant charging piles is dynamically adjusted based on the updated scenario set and preset triggering conditions related to demand fluctuations or equipment failures.

[0009] Furthermore, in step S1, the collected charging demand data includes at least: Real-time operating parameters of the regional power grid obtained through load sensors, user travel trajectories and discrete charging behavior records obtained through vehicle terminals or traffic management systems, and regional development dynamic data obtained through planning department interfaces. The charging pile operation data includes at least: The charging pile's core component operating parameters are obtained through testing sensors, and historical fault records are obtained through the operation and maintenance system.

[0010] Furthermore, the collection of charging demand data and charging pile operation data in step S1 specifically includes: Demand data collection: Real-time operating parameters of the regional power grid are obtained through load sensors; user behavior data, including travel trajectories and charging durations, are obtained through application programming interfaces (APIs) of vehicle terminals or traffic management platforms; and regional development dynamic data are obtained through planning data interfaces. Equipment data acquisition: The operating parameters of the core components are collected through test sensors deployed on the charging piles; historical fault records and related information are obtained through the charging pile operation and maintenance management system. Auxiliary data acquisition: Obtain the power grid load redundancy threshold through the power grid dispatching system interface, and obtain extreme weather early warning information through the meteorological data interface.

[0011] Further, the step S1 of obtaining the demand confidence interval based on the demand forecasting model specifically includes: Model training: The Bayesian neural network is trained using preprocessed historical charging demand data as supervision labels; Interval prediction: Based on the trained model, the feature data of the prospective period is used as input to output the predicted interval of charging demand at different confidence levels, and the uncertainty of demand fluctuation is represented in the form of a probability distribution. Scenario segmentation: Using the Monte Carlo simulation method, multiple demand scenarios covering different fluctuation intensities are sampled from the probability distribution to form an instantiation of the demand confidence interval.

[0012] Further, generating the joint uncertainty scenario set in step S1 specifically includes: Coupling operation: The demand fluctuation probability distribution represented by the demand confidence interval is coupled with the equipment failure probability distribution represented by the failure confidence interval to construct a joint probability model representing two-dimensional uncertainty; Scenario generation: Based on the aforementioned dual-dimensional joint probability distribution model, multiple joint scenarios with definite probabilities are generated through sampling calculation; wherein, each joint scenario is jointly defined by a combination of demand fluctuation range at a specific confidence level and equipment failure at a specific probability. Weighting and Constraint Extraction: Assign a weight to each generated joint scenario, which is equal to the probability of occurrence of the scenario in the two-dimensional joint probability distribution model; and identify extreme joint scenarios with a probability of occurrence below a preset threshold as hard constraints that must be satisfied in subsequent optimization solutions.

[0013] Furthermore, the improved non-dominated sorting genetic algorithm in step S2 dynamically allocates weights based on the intensity of demand fluctuations and the degree of equipment aging, specifically including: Feature quantification: Based on the joint uncertainty scenario set, calculate the scenario intensity index that characterizes the overall demand fluctuation range, and calculate the regional aging index that characterizes the overall health status of the equipment based on the fault confidence interval. Dynamic weight generation: In each iteration of the algorithm, based on the real-time values ​​of the scene intensity index and the area aging index, a combination of weight coefficients is dynamically generated through preset mapping rules to balance the relationship between service continuity objectives and cost reliability objectives. Weighting Application: The dynamically generated weight coefficients are combined and applied to the solution process of the multi-objective optimization function, so that the search direction of the algorithm can adaptively tend to alleviate the more prominent risk dimension in the current uncertainty features.

[0014] Furthermore, the construction and solution of the multi-objective optimization function in step S2 specifically includes: Function Construction: The constructed multi-objective optimization function is expressed as follows: ,in, To improve service satisfaction, For total lifecycle cost, For resource utilization, , , These are the preset target weights for service satisfaction rate, cost, and resource utilization rate; Dynamic coupling of coefficients: The values ​​of the service satisfaction rate target weight and the cost target weight are not fixed constants, but are coupled with the dynamically generated weight coefficient combination, so that the core trade-off relationship of the multi-objective optimization function can be adjusted in real time according to the system risk characteristics represented by the joint uncertainty scenario set generated in step S1. Solution set mapping output: The multi-objective optimization function is optimized by the improved non-dominated sorting genetic algorithm, and a scheme is selected from the final non-dominated solution set. The scheme directly outputs the redundant stub configuration parameters corresponding to the current weight tendency.

[0015] Furthermore, the determination of the number of redundant charging piles in step S2 is specifically achieved in the following way: Dynamic parameter acquisition: The key input parameters in the redundant configuration model—demand fluctuation rise rate and regional average failure probability—are taken in real time from specific statistical values ​​in the latest output of the demand confidence interval and the failure confidence interval in step S1. Redundancy requirement calculation: The dynamically acquired parameters are substituted into the redundancy configuration model. The logic of the redundancy configuration model is configured as follows: when the demand fluctuation rise rate increases significantly, the calculation result tends to increase the redundancy resources to prevent demand peaks; when the regional average failure probability increases significantly, the calculation result tends to increase the redundancy resources to resist equipment cluster failures. Strategy execution: The calculated total redundancy requirement, combined with the layout position obtained by the multi-objective optimization function, constitutes the configuration scheme.

[0016] Furthermore, the real-time data acquisition in step S3 to drive the closed-loop iteration of the model and solution specifically includes: Feedback data collection: After the configuration scheme is implemented, the real-time operation feedback data of the target area is continuously acquired through the deployed data collection node network. The feedback data includes the actual charging demand sequence and the charging pile operation status sequence. Feedback feature processing: The real-time operation feedback data is processed in real time to extract dynamic demand features and dynamic equipment health features for model updates; Model Iteration Driven: The extracted dynamic demand features and dynamic equipment health features are used as incremental training data and input into the demand prediction model and fault prediction model in step S1 to drive the update of the demand confidence interval, the fault confidence interval and the joint uncertainty scenario set.

[0017] Furthermore, the dynamic iterative optimization in step S3 is implemented through the following closed-loop process: Model and scenario set update: Based on the real-time operation feedback data, the demand prediction model and the fault prediction model are retrained in a rolling manner at a period not exceeding the preset update time, and the demand confidence interval, the fault confidence interval and the joint uncertainty scenario set are updated accordingly. Layout scheme re-optimization trigger: When the updated joint uncertainty scenario set indicates that the change in system risk characteristics reaches a set threshold, the robust redundancy layout solution process of step S2 is automatically triggered, with the updated scenario set as input. Dynamic deployment of the solution: The new configuration solution obtained by resolving the solution is compared with the current running solution, and an executable set of adjustment instructions is generated to complete the seamless switching and dynamic deployment from the old solution to the new solution.

[0018] The beneficial effects of this invention are: Compared with the prior art, the present invention has at least the following beneficial effects: 1. Achieving coordinated quantification of demand uncertainty and equipment uncertainty, significantly improving system robustness: This invention breaks through the limitation of existing technologies that only plan for a single uncertain factor by simultaneously modeling the uncertainty of charging demand fluctuations and charging pile operation reliability, and conducting coupled analysis under a unified framework. This enables the layout scheme to effectively cope with complex operating scenarios with multiple adverse factors superimposed, and improves the service stability and risk resistance of the charging system under extreme conditions.

[0019] 2. Redundancy layout optimization based on uncertain scenarios, balancing service assurance and resource efficiency: This invention optimizes redundancy configuration based on multiple uncertain joint scenarios, so that the setting of redundant resources no longer depends on fixed reservation strategies, but matches with changes in demand and equipment operating status. This ensures the continuity of charging services under high load or abnormal conditions, while reducing resource idleness under normal operating conditions, and achieves a synergistic improvement in service capacity and resource utilization efficiency.

[0020] 3. Improve the long-term adaptability of the layout scheme through dynamic iteration mechanism: The present invention introduces a rolling update mechanism based on real-time operation data, which can continuously correct the prediction results and layout scheme as demand changes and equipment status evolves, so that the charging pile configuration scheme has the ability to adapt to the evolution of time, avoiding the problem of traditional one-time planning gradually becoming ineffective in long-term operation.

[0021] 4. Complete methodology and high feasibility for engineering implementation: The demand forecasting, equipment status assessment and optimization solution processes involved in this invention can all be realized based on existing data acquisition conditions and mature calculation methods. It can be deployed in harmony with existing charging pile systems, monitoring systems and operation and maintenance platforms without the need for large-scale transformation of existing infrastructure, and has good feasibility for engineering implementation.

[0022] 5. Comprehensive improvement of economic and social benefits: By improving the rationality of charging facility layout and operational reliability, this invention can reduce the waste of construction and maintenance costs caused by insufficient or excessive configuration, reduce user charging waiting time and service interruption risks, and at the same time help improve the user experience of electric vehicles and the overall operating efficiency of related energy systems, thus having significant economic and social value. Attached Figure Description

[0023] Figure 1 This is a flowchart of the steps in the electric vehicle charging pile demand prediction and layout optimization method of the present invention. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0025] Example 1, refer to Figure 1 This is the first embodiment of the present invention. This embodiment provides a method for predicting the demand and optimizing the layout of electric vehicle charging piles, which can achieve a robust balance between charging service continuity, resource utilization and total life cycle cost in complex scenarios with multiple uncertainties.

[0026] The following detailed explanation of the electric vehicle charging pile demand prediction and layout optimization method of the present invention, based on the actual application scenario in City A, is provided. This embodiment is only used to explain the technical solution of the present invention and does not constitute a limitation on the scope of protection.

[0027] I. Description of Implementation Environment and Application Scenarios; The area of ​​City A is approximately 12 square kilometers, with a current resident population of about 150,000 and about 8,000 electric vehicles. There are 120 charging stations already installed in the area, including 60 fast charging stations and 60 slow charging stations, located in residential parking lots, commercial complexes, and roadside parking spaces.

[0028] According to regional planning information, City A will construct two large office building clusters and add a new subway extension station, which is expected to increase the permanent population by approximately 30,000. The annual growth rate of electric vehicle ownership is expected to be approximately 20%. The existing charging piles have an average operating life of about 3 years, and historical maintenance data shows that level 2 faults are the most common, occurring an average of 8 times per year, while level 3 faults occur an average of 1 time per year.

[0029] Against this backdrop, regional charging demand shows a clear upward trend and exhibits significant fluctuations during morning and evening rush hours and extreme weather conditions. Meanwhile, the risk of failure due to equipment aging is gradually increasing. There is an urgent need for a method for predicting and optimizing the layout of charging piles that can simultaneously address the uncertainties of both demand fluctuations and equipment failures.

[0030] II. Overall working principle of Example 1; In this embodiment, the overall working principle of the method of the present invention is as follows: By collecting charging demand data and charging pile operation data within the region, a demand prediction model and an equipment failure prediction model are constructed respectively. The fluctuations in charging demand and the failure of charging piles in the future are probabilistically quantified to form a demand confidence interval and a failure confidence interval. On this basis, the two types of confidence intervals are coupled to generate a set of joint uncertainty scenarios that include multiple joint operating states.

[0031] Subsequently, using a set of joint uncertainty scenarios as constraints, a multi-objective optimization model was constructed with charging service satisfaction rate, resource utilization rate, and total life cycle cost as objectives. The optimal redundancy layout scheme, which includes hot backup redundant piles and cold backup redundant piles, was obtained by solving the model through an improved non-dominated sorting genetic algorithm.

[0032] After the implementation of the solution, the demand prediction model and fault prediction model are continuously updated by collecting real-time demand data and equipment operation data, and the configuration scheme of redundant charging piles is dynamically adjusted according to preset trigger conditions, so as to achieve continuous optimization and long-term adaptation of the charging facility layout.

[0033] III. Implementation Steps; (a) Step S1: Two-dimensional uncertainty quantification; 1. Data collection; In this embodiment, the data acquisition module is deployed in City A area, specifically including: 1) Demand-related data acquisition devices and environment; The power grid load sensor collected real-time power data of the regional power grid. The collection results showed that the peak load of the region was about 1200kW, and the power grid overload threshold was set at 1500kW. We obtained approximately 50,000 daily travel trajectory data points in the region through the local interface of Gaode Maps and ride-hailing platform terminals, and extracted that the charging demand during the morning peak (7:00–9:00) accounted for 35% and the fast charging preference rate was 60%. Time-series data on newly built office buildings to be delivered in 2025 and subway stations to be put into use in 2026 were obtained through the urban planning management system.

[0034] 2) Equipment operation data acquisition device and environment; Operating parameters of 120 charging piles were collected using operating status sensors that conform to the G01M classification, including charging module temperature, cumulative number of charging times, and voltage fluctuation frequency. The highest annual temperature of the charging module was 45℃, the average number of cumulative charging times per pile was 8,000, and the voltage fluctuation frequency was about 3 times per month. Historical fault records are obtained through the operation and maintenance management system. There are an average of 8 level 2 faults per year, with a single repair time of 4-6 hours. There is an average of 1 level 3 fault per year, with a repair time of about 10 hours.

[0035] 3) Auxiliary operation data; Set the power grid load redundancy threshold to 300kW; Energy storage devices can participate in peak shaving and valley filling when the remaining power is not less than 50%. The region experiences an average of 15 days of high-temperature warnings in summer and 5 days of cold wave warnings in winter.

[0036] 2. Data preprocessing; The collected data was cleaned and standardized to remove outliers. Specifically, a single detected instantaneous load of 1800kW was identified as a sensor malfunction and removed. At the same time, the cumulative duration of high-temperature operation of the equipment (average 1200 hours) and the frequency of voltage fluctuations (approximately 0.1 times / day) were extracted as equipment fault characteristics.

[0037] 3. Two-dimensional uncertainty quantification and coupling; 1) Quantifying demand fluctuations; A Bayesian neural network model is trained based on the demand feature set to output the probability distribution of charging demand in future periods. At a 95% confidence level, the current morning peak charging demand is in the range of 300–450 times / hour; after the new office buildings are put into use in 2025, the peak demand will rise to 400–550 times / hour; under high temperature warning conditions, the demand will further increase by 20%.

[0038] 2) Equipment fault quantification; Based on the equipment feature set, a fault prediction model was trained, and the average fault probability of charging piles in the next month was 12%, the probability of two charging piles failing at the same time was 3.5%, and the probability of a level 3 fault was 0.8%, and corresponding fault confidence intervals were formed for each.

[0039] 3) Constructing scenarios with joint uncertainty; The demand confidence interval and the fault confidence interval are coupled to generate a set of joint uncertainty scenarios, including a scenario of "demand peak of 550 times / hour and two level-two faults occurring simultaneously" with an occurrence probability of 3.5%, and a scenario of "demand peak of 450 times / hour and one level-three fault occurring" with an occurrence probability of 0.8%.

[0040] (ii) Step S2: Solving for robust redundancy layout; 1. Setting constraints and objective function; In this embodiment, the following constraint is set: the charging service fulfillment rate is not less than 98% under extreme scenarios; The probability of grid overload is no higher than 3%; the response time of hot backup redundant charging piles is no more than 3 minutes; the utilization rate of charging pile resources is no less than 75% under normal operating conditions; and minimizing the total life cycle cost is one of the optimization objectives.

[0041] 2. Multi-objective optimization solution; Based on the characteristics of regional demand growth, the weighting coefficient for demand uncertainty is set to 0.6, and the weighting coefficient for equipment failure uncertainty is set to 0.4. An improved non-dominated sorting genetic algorithm is used to solve the multi-objective optimization model.

[0042] In the calculation, it is assumed that the service capacity of a single fast charging station is 6 times / hour, and the service capacity of a single slow charging station is 2 times / hour. The theoretical service capacity of 60 fast charging stations and 60 slow charging stations is 480 times / hour. Under the condition of a peak demand of 660 times / hour and a decrease in the number of available charging stations due to a 12% failure rate, the redundancy service capacity gap is calculated.

[0043] The final redundancy configuration plan is as follows: 10 hot backup redundant charging piles, all of which are fast charging piles, deployed in commercial complexes and around subway stations; and 8 cold backup redundant charging piles, with reserved installation locations near the power grid load redundancy nodes.

[0044] (III) Step S3: Dynamic iterative optimization; Six months after the system was implemented, it was detected that the overall failure rate of the charging piles increased to 14% due to the increased operating time caused by high summer temperatures, reaching the preset trigger condition. Based on this, the system automatically adjusted the redundancy configuration scheme, quickly installing and putting into operation two cold backup redundant piles, while switching the hot backup redundant piles to active load-sharing mode.

[0045] After the adjustment, the charging service satisfaction rate is 96% under normal operating conditions, and 98.5% under extreme conditions.

[0046] IV. Simulation Verification and Implementation Results; By constructing a digital twin model of the Jia City area, a simulation verification was conducted on the extreme scenario of "550 times / hour of demand during the morning peak after the office buildings are delivered in 2025, and 3 fast charging piles malfunctioning". The results showed that the charging service satisfaction rate was 98.2%; the average waiting time for users was 8 minutes; the peak grid load was 1450kW, which did not exceed the 1500kW overload threshold; and the resource utilization rate reached 78%.

[0047] Therefore, it can be seen that the method of this embodiment can significantly improve the service reliability, resource utilization efficiency and overall operational stability of the charging system under complex and uncertain operating conditions.

[0048] Example 2 is the second embodiment of the present invention.

[0049] I. Overall technical principles of Example 2; Based on the engineering application described in Example 1, this second embodiment further explains the core implementation mechanism of steps S1 to S2 of the present invention from the perspective of algorithm model and system risk modeling. It focuses on the generation method of joint uncertainty scenario set and its decisive role in robust redundancy layout optimization.

[0050] The core idea of ​​this embodiment is: Both the uncertainty of charging demand fluctuations and the uncertainty of charging pile equipment failures are modeled in the form of probability distributions and confidence intervals, and structuredly coupled in a unified probability space to generate a set of joint uncertainty scenarios containing multiple combinations of "demand state - failure state" and their probabilities of occurrence. Subsequently, in step S2, this set of joint uncertainty scenarios is used as the sole constraint input source for multi-objective optimization, so that the layout scheme has the ability to cope with complex and extreme operational risks in the design stage.

[0051] II. The structure, construction steps, and mathematical expression of the demand forecasting model; (a) The structure and construction steps of the demand forecasting model; In this embodiment, the demand forecasting model is built based on a Bayesian neural network to characterize the stochastic fluctuations in charging demand across time and space. The model construction steps include: 1. Collect and preprocess a demand-related feature set, which includes at least regional power grid load characteristics, user travel trajectory characteristics, and regional development dynamic characteristics; 2. Using historical charging demand data as supervision labels, the Bayesian neural network is trained so that the network weights and bias parameters follow the posterior probability distribution. 3. During the prediction phase, the probability distribution of charging demand within the future prediction time window is obtained through multiple random sampling methods; 4. Based on the probability distribution, output the demand confidence interval at different confidence levels to characterize the uncertainty of demand fluctuations.

[0052] (ii) Mathematical expression of the demand forecasting model; In this embodiment, the probability intensity function of the charging demand per unit time within the prediction time window is defined as: ; This formula comprehensively characterizes the asymmetric distribution characteristics, periodic fluctuation characteristics, and long-tail effects caused by extreme events of demand by introducing the Gamma function, Bessel function, error function, and Riemann Zeta function.

[0053] (iii) A complete explanation of formula characters; in, This represents the probability intensity value for predicting charging demand per unit time within a predicted time window. To continuously represent potential demand variables, Here, Γ is the demand distribution shape adjustment parameter, and Γ is the Gamma function used to describe the heavy-tailed characteristics of the demand distribution. Here, ln is the demand fluctuation amplification factor, and ln is the natural logarithm function. for The first-order Bessel function of the first kind is used to characterize periodic perturbations in travel behavior. For periodic modulation parameters, It is an exponential function. Higher-order attenuation parameters are used to suppress abnormally large demands. The error function is used to calculate the cumulative probability that demand exceeds a given confidence center. For the confidence center parameters of the demand, For Riemann Zeta function, This is a parameter for adjusting the frequency of extreme events.

[0054] (iv) The technical effectiveness of the demand forecasting model; The demand forecasting model described above can be used to obtain the demand confidence interval expressed in the form of a probability distribution, thereby avoiding the problem of underestimating demand risk caused by relying on a single forecast value and providing a basic input for subsequent joint uncertainty modeling.

[0055] III. Structure, construction steps, and mathematical expression of the fault prediction model; (I) The structure and construction steps of the fault prediction model; Structure of the Fault Prediction Model: The fault prediction and health management model adopts an ensemble learning structure, preferably using an improved random forest model. This model includes an input feature layer, a feature mapping and tree model ensemble layer, and a probability output layer.

[0056] The input feature layer receives preprocessed equipment operation features. The feature mapping and tree model integration layer consists of multiple decision trees, each establishing a nonlinear mapping relationship between equipment operating status and fault events based on different training sample subsets and feature subsets. The probability output layer aggregates the prediction results of each decision tree and outputs the fault probability, fault level, and fault duration interval of the target charging pile within the prediction time window.

[0057] In this embodiment, the fault prediction model uses an improved random forest to model the operating characteristics of charging piles. The construction steps include: 1. Collect and preprocess the equipment operation feature set, which includes at least the component aging rate, voltage fluctuation frequency, cumulative high-temperature operation time, and vibration frequency; 2. Use historical fault records as supervision labels to train the fault prediction model; 3. The probability distribution of the failure probability and failure duration of the output charging pile within the predicted time window; 4. Based on the output results, form the fault confidence interval and distinguish different fault levels.

[0058] How to construct input features: The model input data originates from the equipment operation data acquisition stage and includes at least the following characteristics: Equipment runtime, cumulative charging times, module operating temperature, cumulative high-temperature operation time, voltage fluctuation frequency, voltage fluctuation amplitude, current stability index, component vibration frequency, historical fault types, historical fault levels, historical repair time, operation and maintenance records, and component replacement records.

[0059] Data of different dimensions are first cleaned, normalized, and feature-encoded to form a unified-dimensional device feature vector. Normalization maps data of different dimensions, such as temperature, runtime, and number of charging cycles, to the same numerical range, facilitating model training. Categorical variables, such as fault type, component category, and maintenance method, can be processed using one-hot encoding or ordinal encoding.

[0060] How to construct training samples: The training samples are constructed according to a time window approach. The operational characteristics of each charging pile within a certain historical time period are used as input samples, and whether a fault occurs within a preset prediction window after that time period, the level of the fault, and the duration of the fault are used as supervision labels to form a supervised learning dataset.

[0061] The predicted labels include at least three categories: The first category is the fault occurrence label, which is used to characterize whether a fault has occurred within the prediction window; The second category is fault level labels, which are used to characterize level one, level two, or level three faults. The third category is fault duration labels, which are used to characterize the duration of fault repair and can be represented by continuous numerical values ​​or intervals.

[0062] The fault severity levels are classified according to the following standards: Level 1 faults have a repair time of no more than 2 hours, Level 2 faults have a repair time of more than 2 hours but no more than 8 hours, and Level 3 faults have a repair time of more than 8 hours.

[0063] Model training process: The model training process includes four stages: sample extraction, decision tree generation, model integration, and parameter calibration.

[0064] First, multiple training subsets are generated from the historical training dataset using sampling with replacement, with each subset used to train a decision tree. Second, during the node splitting process of each decision tree, a subset of features is randomly selected from all features as candidate splitting features, and the optimal splitting feature is selected based on information gain, information gain ratio, or Gini index, thus forming the tree structure. Subsequently, depth constraints or pruning are applied to the generated decision trees to reduce the risk of overfitting. Finally, multiple decision trees are integrated into a random forest model, and the model parameters are determined through cross-validation.

[0065] The optimal parameter range can be set as follows: the number of decision trees is 50 to 200, the maximum depth of each decision tree is 5 to 15 layers, the number of features selected in each node split is 30% to 70% of the total number of features, and the minimum number of leaf node samples is 2 to 20. These parameters can be adjusted according to the data volume of charging pile clusters of different regions and sizes.

[0066] Model output method: After training, the model can predict the failure risk of any target charging pile within the prediction time window. During prediction, the equipment operation characteristics collected at the current moment are input into the model, each decision tree outputs the corresponding failure prediction result, and then the results are aggregated through averaging, weighted averaging, or voting to obtain the final output.

[0067] The output should include at least: the failure probability of a single charging pile; the probability distribution of the failure level of a single charging pile within the prediction window; and the failure duration interval of a single charging pile. The average failure probability of a cluster of charging piles within the region; the combined probability of multiple charging piles failing simultaneously.

[0068] The regional average failure probability can be obtained by taking the average failure probability of all charging piles in the region; the combined probability of multiple charging piles failing simultaneously can be obtained by taking historical joint statistics, conditional probability statistics, or combined probability estimation.

[0069] How the fault confidence interval is formed: After obtaining the fault prediction results for individual devices and regional device clusters, fault confidence intervals can be further formed. Fault confidence intervals include fault probability confidence intervals and fault duration confidence intervals.

[0070] The failure probability confidence interval characterizes the fluctuation range of the failure occurrence probability within the prediction window, while the failure duration confidence interval characterizes the possible range of failure repair duration. Both reflect equipment-side uncertainty and serve as one of the inputs for constructing a joint uncertainty scenario set.

[0071] For example, for a charging pile cluster in a certain area that has been in operation for 3 years, the model can output that the average failure probability in the next month is 12%, the probability of two charging piles failing at the same time is 3.5%, the probability of a level 3 failure is no higher than 1%, the repair time for a level 2 failure is 4 to 6 hours, and the repair time for a level 3 failure is greater than 8 hours.

[0072] Online model update mechanism: Combined with a dynamic iterative optimization process, the fault prediction and health management model supports both periodic and incremental updates. The update cycle does not exceed 24 hours.

[0073] When the cumulative average failure probability in the region increases to more than 8%, or the failure frequency of a certain key component is significantly higher than the historical average, the latest collected equipment dynamic data, failure warning information, maintenance records and component replacement information are added to the training sample set, and the model is retrained or incremental updates are performed to obtain a new failure probability distribution and failure duration range.

[0074] To ensure the model can continuously reflect equipment aging and changes in the operating environment, a sliding time window approach can be used to manage training data. The window retains recent, valid samples while discarding outdated and less representative data, thereby improving the model's adaptability to the current operating conditions.

[0075] Through the above structure and training mechanism, the fault prediction and health management model can achieve the following technical effects: First, it can simultaneously output the failure probability and the failure duration range, making the uncertainty quantification results on the equipment side more complete; Secondly, it can process multi-source heterogeneous operating data, improving the accuracy and stability of fault prediction; Third, it can extend from single charging piles to the fault risk assessment of regional charging pile clusters, providing reliable input for the generation of joint uncertainty scenario sets; Fourth, by using periodic and incremental update mechanisms, the prediction distortion caused by model decay is reduced, thereby improving the timeliness and reliability of subsequent redundant layout optimization.

[0076] (ii) Mathematical expression of the fault prediction model; In this embodiment, the charging pile fault risk intensity function within the prediction time window is defined as: ; This formula characterizes the cumulative effect of equipment degradation through the numerator and introduces environmental stress and multiple fault correction factors into the denominator to normalize the model of fault risk.

[0077] (iii) A complete explanation of formula characters; To predict the intensity of charging pile failure risk within a time window, For continuous variable such as equipment runtime, To predict the length of the time window, The m-th order modified Bessel function of the second kind is used to describe the coupling effect between vibration and thermal stress. For degradation rate modulation parameters, For nonlinear degradation amplification parameters, For the Gamma function, Let be the prior shape parameter of the fault, and exp be the exponential function. The degradation attenuation coefficient, The exponential integral function is used to describe the accumulation process of latent faults. Environmental stress coefficient, For Riemann Zeta function, Correction parameters for multiple fault superposition.

[0078] (iv) The technical effectiveness of the fault prediction model; This model can identify the risk level of different levels of equipment failure within the prediction time window in advance, providing a reliable supply-side uncertainty input for the subsequent construction of joint uncertainty scenarios.

[0079] IV. Generation methods and structural features of joint uncertainty scenario sets; (i) The idea of ​​generating a set of joint uncertainty scenarios; In this embodiment, step S1 does not process the demand forecast result and the fault forecast result separately, but treats both as uncertain outputs existing in the form of probability distributions, and performs active and structured coupling modeling based on this.

[0080] Demand confidence intervals reflect the uncertainty of demand-side behavior, while fault confidence intervals reflect the uncertainty of equipment status on the supply side. The two differ significantly in their physical mechanisms and statistical characteristics, but they both affect the charging system in actual operation.

[0081] (ii) Steps for constructing a set of joint uncertainty scenarios; 1. Map the demand confidence interval and the fault confidence interval to a unified probability space; 2. Establish a nonlinear coupling relationship between demand uncertainty and fault uncertainty within a unified probability space; 3. Generate multiple joint uncertainty scenarios through random sampling or Monte Carlo simulation; 4. Assign corresponding occurrence probability weights to each joint uncertainty scenario; 5. Identify joint scenarios with a probability of occurrence below a preset threshold but significant risk consequences as extreme joint uncertainty scenarios.

[0082] Each joint uncertainty scenario includes at least: demand fluctuation range, equipment failure combination state, and corresponding joint probability of occurrence.

[0083] V. The decisive role of the joint uncertainty scenario set in step S2; In this embodiment, the robust redundancy layout optimization in step S2 no longer directly uses the demand forecast results or the average equipment failure rate as constraints, but only uses the joint uncertainty scenario set as the constraint input source for multi-objective optimization solution.

[0084] Specifically, constraints such as charging service fulfillment rate, grid overload risk, and redundancy response timeliness are all evaluated under the composite operating state defined by the joint uncertainty scenario. A redundancy layout scheme is only deemed feasible if it meets the constraints in multiple joint uncertainty scenarios.

[0085] This optimization method enables the layout scheme to inherently possess the ability to cope with the "overlapping occurrence of peak demand and equipment failure" during the design phase.

[0086] Structure and inference rules of the two-dimensional joint probability distribution model: (a) Model structure; The two-dimensional joint probability distribution model outputs the demand forecasting model. With the output of the fault prediction model As input, a joint probability space is constructed through a nonlinear coupling function. Its structure includes a demand edge probability mapping layer, a fault edge probability mapping layer, a correlation coupling layer, and a scene sampling layer.

[0087] (II) Construction Steps; right and Normalization is performed separately to form the probability distribution of demand edge and the probability distribution of failure edge; By introducing nonlinear correlation parameters, a demand-fault coupling relationship is constructed. Joint samples were generated using Monte Carlo simulation. Calculate the probability of occurrence for each joint sample and assign weights accordingly; The probability of occurrence is lower than a preset threshold (e.g., 1%) and simultaneously meets the following conditions: and Joint samples that all exceed the risk threshold are identified as hard constraint scenarios that must be met.

[0088] (iii) Rules of reasoning logic; when and When both are low, it corresponds to the normal operating scenario; when higher and When the level is low, it corresponds to demand-driven scenarios; when lower and When the value is higher, it corresponds to a scenario dominated by equipment failure; when and At the same time, when the level is high, it corresponds to the extreme joint uncertainty scenario, which is given priority to be included in the mandatory constraint conditions of robust redundancy layout optimization.

[0089] VI. The comprehensive technical effects and creative embodiment of Example 2; The method described in Example 2 achieves at least the following technical effects: The approach has shifted from planning for single uncertainties to system-level risk planning based on joint uncertainty scenarios. By replacing deterministic constraints with probabilistic constraints, the statistical reliability of the layout scheme is improved. A new decision-making object—the joint uncertainty scenario set—is formed, and its generation and usage methods are different from existing technologies. Significantly improves the robustness and service support capabilities of the charging system under extreme operating conditions.

[0090] Example 3 is the third embodiment of the present invention.

[0091] I. Overall technical concept and working principle of Example 3; Based on the completion of the joint uncertainty scenario set described in Example 2, this third embodiment further optimizes the solution process of the redundant charging pile layout scheme in step S2.

[0092] The core technical idea of ​​this embodiment is: Instead of using demand forecasting or fault prediction results as independent input parameters, the system operating state represented by the joint uncertainty scenario set is used as the only source of uncertainty. An improved non-dominated sorting genetic algorithm is driven by a dynamic weight adjustment mechanism to complete the robust optimization solution of redundant configuration.

[0093] This approach enables the optimization algorithm to adaptively respond to the relative dominance of demand-side risks and equipment-side risks under different joint scenarios during the search of the solution space, thereby avoiding the planning rigidity problem caused by fixed weights in traditional methods.

[0094] II. Multi-objective optimization constraints and threshold settings; (a) Hard constraints; In this embodiment, all candidate redundant layout schemes must satisfy the following hard constraints: Under extreme joint uncertainty scenarios, the charging service satisfaction rate shall not be less than 98%, where the service satisfaction rate is defined as the ratio of the actual number of charging completed to the number of demanded times within the corresponding time window; In any combined scenario, the overload probability of the regional power grid is no higher than 3%, and the overload probability is calculated statistically based on real-time load data collected by load sensors. When a normal charging pile experiences a level 2 or higher fault, the fault replacement response time of the hot backup redundant charging pile shall not exceed 3 minutes. Once the activation conditions are triggered, the cold backup redundant charging pile will be installed and put into operation within 48 hours.

[0095] (ii) Soft constraints; Simultaneously, the following soft constraints are set to improve the long-term economic efficiency and operational efficiency of the system: Under normal joint scenarios, the overall utilization rate of charging piles in the area shall not be less than 75%, where the utilization rate is defined as the ratio of actual charging time to equipment available time. Minimize total lifecycle costs, which include construction costs, operation and maintenance costs, and service loss costs due to failures. The risk output update cycle of the fault prediction model does not exceed the preset update time (24 hours), and the redundant configuration parameters can be dynamically adjusted according to the change of fault probability.

[0096] III. Risk mapping and weight generation model driven by joint uncertainty; (a) Construction of the risk mapping function; To avoid duplication with the mathematical structures of demand forecasting and failure prediction models, this embodiment maps the joint uncertainty scenario set into two intermediate risk indicators, which are used to characterize demand-side pressure and equipment-side pressure, respectively: ; ; in, The number of samples for the required scenarios. Index for demand scenarios, This represents the upper limit of demand in the i-th demand scenario. To correspond to the lower limit of demand, Let N be the average demand, N be the number of charging stations in the area, and j be the index of the charging station. Let j be the equivalent operating time of the j-th charging pile. The failure rate per unit time output by the failure prediction model. To assess the intensity of demand risk, This represents the intensity of equipment failure risk.

[0097] In step S1, the demand confidence intervals at different confidence levels have been obtained through the Bayesian neural network model. These confidence intervals exist in the form of upper demand limit, lower demand limit, and their statistical mean.

[0098] The first formula above does not re-predict demand, but rather quantifies the uncertainty intensity of demand fluctuations a second time based on the obtained demand confidence interval results. The formula uses the ratio of the demand interval width to the mean demand to characterize the relative fluctuation intensity under a single demand scenario, and by summing and averaging multiple demand scenarios, it maps discrete demand uncertainty results into a continuous demand risk intensity. The introduction of the logarithmic function causes the risk indicator to exhibit non-linear growth characteristics when the demand fluctuation range expands sharply, thereby avoiding the unreasonable linear amplification of extreme values ​​in the subsequent optimization process.

[0099] This formula solves the following problems in existing technologies: Existing technologies usually only use "peak demand" or "average demand" as the basis for planning, which cannot reflect the differences between different demand confidence intervals, nor can they distinguish between the two types of risk characteristics: "stable high demand" and "drastic fluctuation demand".

[0100] This demand-risk mapping formula transforms the uncertainty on the demand side in a joint uncertainty scenario from an interval form into a continuous risk variable that can participate in optimization calculations, providing a unified and computable input basis for subsequent dynamic weight generation and redundancy configuration.

[0101] In this embodiment, the prediction result is not replaced, but is used as input to construct the overall risk intensity on the device side.

[0102] The second formula above obtains the cumulative risk of a single charging pile failing within its future operating cycle by exponentially mapping the equivalent operating time of a single charging pile to the failure rate per unit time. Then, by averaging across all charging piles in the region, the regional-level equipment failure risk intensity is formed. The introduction of the exponential function makes the impact of equipment aging on risk exhibit the characteristics of "slow in the early stage and accelerated in the later stage", which is more in line with the actual equipment failure mechanism.

[0103] This formula solves the following technical problem: Existing technologies typically only use "average failure rate" or "historical number of failures" as reliability indicators, which cannot reflect the coupling relationship between equipment uptime differences and failure probability, and easily underestimate the systemic risks brought about by aging equipment clusters.

[0104] By using this equipment failure risk mapping formula, equipment-side uncertainties can participate in subsequent optimization in the form of regional risk intensity, thereby avoiding static defense against single-point failures in redundant layouts.

[0105] (ii) Weight generation and normalization mechanism; After obtaining the intensity of demand risk and the intensity of failure risk, the weight generation function is constructed as follows: ; in, As a weight for the service satisfaction rate target, As a weight for cost targets, The target weight for resource utilization rate.

[0106] The above formula normalizes the intensity of demand risk and the intensity of equipment failure risk, so that the weighting coefficients are no longer preset constants, but are generated in real time by the joint uncertainty scenario set.

[0107] Specifically, when the intensity of demand risk dominates, the weight of service satisfaction rate automatically increases; when the intensity of equipment risk dominates, the weights related to cost and reliability automatically increase.

[0108] This formula solves the following technical problem: In existing technologies, multi-objective optimization generally uses fixed weights or manual experience weights, which cannot be adaptively adjusted with changes in uncertain structures, resulting in distortion of optimization results when demand fluctuates drastically or equipment ages intensively.

[0109] IV. Multi-objective optimization functions and their solution methods; In this embodiment, the following multi-objective optimization function is constructed: ,in, To improve service satisfaction, For total lifecycle cost, For resource utilization rate.

[0110] By introducing dynamically generated weight coefficients into the improved non-dominated sorting genetic algorithm, the algorithm can adaptively adjust the search direction under different joint uncertainty scenarios.

[0111] V. Redundant charging pile configuration model and its calculation formula; (a) Redundancy Calculation Model; To improve the feasibility of the project, this embodiment adopts a redundancy quantity calculation model with a clear structure: ; (ii) Explanation of symbols; in, The required number of redundant charging stations. For the peak of regional demand, The upward adjustment rate for demand fluctuations is taken from the maximum upward adjustment value under the 95% confidence level of the demand confidence interval. This refers to the average service capacity of a single charging station. The regional average failure probability is derived from the failure prediction model's forecast for the next month. This refers to the number of existing charging stations that are currently in normal operation.

[0112] The number of redundant charging piles should be dynamically calculated based on the demand fluctuation rate and the regional average failure probability.

[0113] The above formula incorporates both demand-side and equipment-side uncertainties into the redundancy calculation process. The first part ensures that the system has sufficient service capacity in the event of increased demand. The second part corrects the effective supply capacity through failure probability to avoid service gaps when equipment fails. Finally, the number of existing normally operating charging piles is deducted to obtain the actual redundancy required.

[0114] This formula solves the following problem: In existing technologies, the amount of redundancy is usually reserved at a fixed ratio, without taking into account the dynamic changes in demand fluctuations and failure probabilities, which can easily lead to insufficient redundancy or over-construction.

[0115] This redundancy configuration formula enables the redundancy quantity to be dynamically adjusted according to changes in joint uncertainty, ensuring service continuity in extreme joint scenarios and suppressing resource waste in normal scenarios.

[0116] In summary, this third embodiment addresses the technical problem through the following mechanisms: transforming the demand confidence interval and the fault confidence interval into a unified risk intensity variable; dynamically generating optimization weights using the risk intensity variable to drive a multi-objective optimization algorithm; and mapping the optimization results to directly implementable engineering parameters through a clear redundant configuration formula.

[0117] This overall mechanism enables the present invention to no longer rely on static planning assumptions, but instead focuses on joint uncertainty scenarios to achieve robust layout optimization oriented towards real operational risks.

[0118] VI. Principles for Selecting Redundancy Types and Determining Layout Locations; Based on the non-dominated solution set output by the improved non-dominated sorting genetic algorithm, the type and layout location of redundant charging piles are determined, where: Hot-backup redundant charging stations should be prioritized for deployment based on demand and risk intensity. High or equipment failure risk intensity Higher areas are used for rapid replacement and load sharing; The cold backup redundant charging pile is only reserved for installation conditions and is activated when triggered by extreme joint uncertainty scenarios to reduce resource idleness in normal scenarios.

[0119] VII. Technical effects of Example 3; The method described in this embodiment three achieves the following technical effects: The set of joint uncertainty scenarios is transformed into risk inputs that can directly drive optimization algorithms, thus avoiding the lag of traditional static programming; Through risk mapping and weight normalization mechanisms, adaptive adjustment of the target weights is achieved. While meeting high service guarantee thresholds, effectively suppress the excessive expansion of redundant configurations; Improve the robustness and economy of charging infrastructure in complex operating environments.

[0120] Example 4 is the fourth embodiment of the present invention.

[0121] I. Overall working principle of Example 4; This fourth embodiment focuses on step S3, constructing a closed-loop iterative optimization mechanism based on real-time operational feedback data.

[0122] Based on the dual-dimensional uncertainty quantification model, joint uncertainty scenario set, and robust redundancy layout scheme formed in Examples 1-3, this mechanism enables the charging pile layout scheme to continuously adapt to regional demand evolution, equipment status changes, and external environmental disturbances through a closed-loop process of "real-time perception - model update - scenario reconstruction - scheme re-optimization - dynamic deployment".

[0123] Unlike the existing layout approach of "one-time planning and long-term fixation", this embodiment continuously introduces real-world operational data into the model update process, suppressing the problem of prediction model degradation over time and preventing the layout scheme from gradually failing due to scene drift, thereby significantly improving the stability and practicality of the system during long-term operation.

[0124] II. Real-time operation feedback data collection and processing; (a) Feedback data collection mechanism; After the redundant layout scheme is implemented, a data acquisition node network is deployed at substations, large parking lots, charging pile clusters and traffic nodes in the target area. The nodes are preferably devices with edge computing capabilities and are adapted to F16M type device bases for on-site deployment.

[0125] The collected feedback data includes at least the following two categories: The demand feedback data includes real-time charging volume, user waiting time, charging failure records, number of new user registrations, and updates on regional development, such as the handover of new residential areas, the opening of commercial districts, and the commissioning of transportation hubs. The equipment feedback data includes real-time operating parameters of the charging pile, fault warning signals, fault occurrence and repair records, operation and maintenance logs, and information on the replacement of key components.

[0126] The aforementioned feedback data is continuously collected in time series format to form an actual operational feedback dataset.

[0127] (ii) Feedback feature processing; The collected real-time operational feedback data is processed in real time to extract dynamic features for model updates, including: Dynamic demand characteristics, such as the growth rate of charging requests per unit time, the trend of peak period extension, and the rate of change of average user waiting time; Dynamic equipment health characteristics, such as changes in failure frequency, rate of decline in the health of key components, and trends in equipment availability.

[0128] The dynamic features are consistent with the features used for model training in Examples 1 and 2 in terms of data structure, so as to ensure the stability and inheritability of the model iteration process.

[0129] III. Rolling Iterative Updates of the Uncertainty Quantification Model; (a) Model iteration cycle and update conditions; In this embodiment, both the demand forecasting model and the fault prediction model are updated on a rolling basis with a cycle not exceeding 24 hours.

[0130] Specifically: Demand fluctuation model update conditions: The Bayesian neural network model is retrained when any of the following occurs in the demand confidence interval compared to the previous period: the demand fluctuation confidence interval rises by 15% or more; or the demand fluctuation confidence interval falls by 20% or more.

[0131] Fault prediction model update conditions: Incremental training of the fault prediction model is triggered when any of the following conditions occur: the average failure probability of charging piles in the area increases to or exceeds 8%; or the failure frequency of a certain key component is significantly higher than the historical statistical level.

[0132] (ii) Synchronous updating of the joint uncertainty scenario set; After the demand confidence interval and the fault confidence interval are updated, the joint uncertainty scenario set is reconstructed based on the coupling method described in Example 2, so that each joint scenario reflects the latest probability characteristics of the superposition of demand fluctuations and equipment faults under the current real operating state.

[0133] This approach ensures that the scenario set used for subsequent optimizations remains highly consistent with the actual operating environment.

[0134] IV. Re-optimization triggering mechanism for robust redundancy layout schemes; (a) Layout adjustment trigger threshold; In this embodiment, the following explicit layout re-optimization trigger conditions are set: Demand triggering conditions: Demand peak increases by 20% or more in extreme joint scenarios; charging service satisfaction rate is less than 90% in normal joint scenarios.

[0135] Fault triggering conditions: The average fault probability in the area reaches or exceeds 15%; the number of Level 3 faults occurs 2 or more times within a consecutive month.

[0136] Regional development trigger conditions: The number of new permanent residents reaches or exceeds 5,000; a new large-scale commercial complex or transportation hub is put into use.

[0137] When any of the above triggering conditions are met, the system automatically calls the updated joint uncertainty scenario set and re-executes the robust redundancy layout optimization solution process described in Example 3.

[0138] (II) Analysis of Differences Between Schemes and Dynamic Deployment; A difference analysis is performed between the newly obtained redundant layout scheme and the current running scheme to generate an executable set of adjustment instructions.

[0139] The adjustment strategy includes at least the following: switching the hot backup redundant charging piles from "standby mode" to "active load sharing mode"; enabling the rapid installation process of cold backup redundant charging piles; upgrading the power configuration of some normal charging piles; and making minor adjustments to the deployment location of redundant charging piles.

[0140] By using the above methods, a seamless transition from the old to the new layout scheme can be achieved, avoiding significant interference with normal charging services.

[0141] V. Digital twin simulation verification under extreme joint scenarios; After the layout scheme is adjusted, a digital twin of the target area is constructed. The digital twin is adapted to the full-element mapping of multivariate measurement data and is used to simulate and verify the extreme joint scenario of "demand peak and equipment failure superposition".

[0142] The overall structural composition of a digital twin: In this embodiment, the digital twin is a multi-dimensional mapping system of the operating status of the charging infrastructure in the target area. It is not a simple simulation model, but is composed of the following four functional layers working together: Physical entity mapping layer: used for structured modeling of charging piles, grid nodes, energy storage devices and user behavior within the target area; State parameter modeling layer: used to continuously parameterize demand state, equipment state, power grid state and environmental state; Joint Uncertainty Scenario-Driven Layer: Used to map a set of joint uncertainty scenarios to twin runtime inputs; Reasoning and evaluation layer: used to infer the system operation results under a given joint scenario and output verification indicators such as service satisfaction rate and power grid load.

[0143] The above layers are interconnected through data interfaces and model interfaces to form a digital twin system that can reflect the real operating status and support reverse optimization.

[0144] Steps for building a digital twin: (a) Physical entity mapping modeling; First, perform entity-level modeling of the physical objects within the target area, specifically including: Charging pile physical model: Create a corresponding virtual object for each charging pile, and record its type (fast charging / slow charging / redundant pile), rated power, current operating status, fault level and repair time parameters; Power grid node model: Establish a topological relationship model of substations, feeder nodes and key load nodes, and record the rated capacity, voltage level and load redundancy threshold of each node; User behavior entity model: Construct a user arrival process model using historical trajectory data and real-time demand data to describe the distribution of user charging requests at different time periods; Environmental and external event model: Parameterizes external events such as extreme weather, peak hours, and holidays to influence demand and equipment operating status.

[0145] Through the above steps, a set of virtual entities that correspond one-to-one with the actual physical system is formed.

[0146] (ii) State parameter modeling and time synchronization; After completing the entity mapping, the running status of each entity is parameterized and a unified timeline is introduced to keep the digital twin synchronized with the real system.

[0147] The state parameters should include at least: Demand status parameters: charging request intensity, peak demand, and demand fluctuation amplitude per unit time; Equipment status parameters: equipment availability, failure probability, current failure level, and estimated remaining repair time; Power grid status parameters: node load, current, voltage deviation, and overload risk level; Redundancy status parameters: whether the hot backup redundant stub is active, and the available time for the cold backup redundant stub.

[0148] All state parameters are updated with a time granularity of no more than minutes to meet the accuracy requirements of extreme scenario simulation.

[0149] Input methods for scenarios involving joint uncertainty in digital twins: In this embodiment, the operation of the digital twin is not based on a single deterministic input, but rather on a set of joint uncertain scenarios as the driving source.

[0150] Specifically: each joint uncertainty scenario is mapped to a defined set of input conditions, including: Peak demand and demand distribution at a specified confidence level, and equipment failure combinations and corresponding failure durations at a specified probability.

[0151] When the digital twin is running, a single joint scenario is used as the initial condition for a complete simulation cycle. During the simulation cycle, the process of demand arrival and the evolution of equipment state are strictly constrained by the joint scenario, avoiding the separation of demand and failure as independent variables.

[0152] This method enables the twin's operating results to accurately reflect the system behavior under conditions of "overlapping demand peaks and equipment failures".

[0153] The reasoning logic rules of digital twins: (a) Demand-equipment-grid linkage reasoning rules; During the simulation, the digital twin performs state deduction according to the following logical rules: Demand-driven rule: At each time step, based on the demand intensity parameter defined in the joint scenario, generate a corresponding number of charging requests and allocate them to the charging pile entity with the optimal spatial location; Equipment availability determination rules: If the charging pile is in a fault state, no new requests can be assigned to the charging pile within the corresponding time step; if it is a hot backup redundant pile and is in an active state, it is allowed to take over the load immediately. Grid constraint propagation rule: When multiple charging piles are working simultaneously, their superimposed impact on the load of grid nodes is calculated in real time. If the load of a node is close to or exceeds the redundancy threshold, the access of new loads is restricted. Fault evolution rules: For a specified fault combination in the joint scenario, fault occurrence and recovery events are triggered on the simulation time axis according to its fault level and repair duration parameters.

[0154] (II) Indicator Calculation and Constraint Verification Rules; After the simulation is completed, the digital twin calculates the following indicators based on the results: charging service satisfaction rate; Average user waiting time; peak load of regional power grid; utilization rate of charging pile resources.

[0155] The aforementioned metrics are used to compare the current layout with the hard and soft constraints set in Example 4. When any hard constraint is not met, the system automatically marks the current layout as an infeasible solution.

[0156] The reverse coupling mechanism between digital twins and layout optimization: When the digital twin simulation results show that the current layout scheme does not meet the hard constraints under a certain extreme joint scenario, the following reverse coupling process is triggered: The corresponding joint scenario is marked as a high-risk scenario, and its constraint weight in the set of joint uncertainty scenarios is increased; Adjust the weight parameters or constraints of the robust redundancy optimization algorithm in Example 3; The redundant layout optimization solution process is re-executed to generate new candidate layout schemes; The new solution is then input into the digital twin for verification until all hard constraints are met.

[0157] This mechanism enables the digital twin to provide closed-loop correction and enhanced constraints for the layout optimization process.

[0158] By introducing a digital twin with a defined structure, parameters, and inference rules in Embodiment 4, the present invention further achieves the following technical effects: This invention avoids the limitations of relying solely on historical data or static models for verification; identifies potential extreme operational risks before implementation, reducing actual deployment risks; provides verifiable and traceable decision-making basis for dynamic iterative optimization; and enhances the feasibility and credibility of the invention in actual engineering deployments.

[0159] (a) Examples of digital twin simulation scenarios; Including but not limited to the following scenarios: During holidays, demand at highway service areas doubles to 600 times per hour, while three charging stations experience a level-two malfunction, with a repair time of 6 hours. During the evening rush hour in the city's CBD, the demand reached 450 times per hour. At the same time, one charging station experienced a level 3 fault, and the repair time was 12 hours.

[0160] (II) Verification metrics and reverse optimization; The simulation results were validated using the following metrics: charging service satisfaction rate; average user waiting time; peak regional power grid load; and resource utilization rate.

[0161] When the simulation results do not meet the hard constraints (e.g., the service satisfaction rate is less than 98%), the weight coefficients and constraint parameters of the robust optimization algorithm in Example 3 are automatically adjusted in reverse, and the layout scheme is solved again until all constraints are met.

[0162] VI. Technical effects and inventiveness of Example 4; The dynamic iteration mechanism described in this embodiment four achieves the following technical effects: Suppressing model decay and scenario drift: Real-time feedback data drives rolling updates of the model, ensuring that the demand prediction model and the fault prediction model continuously match the actual operating state; Achieving long-term effectiveness of the layout scheme: Redundant layout schemes are no longer a one-time planning result, but can dynamically evolve with regional development and equipment aging; Constructing a complete closed-loop optimization system: forming a closed loop between perception, prediction, optimization and execution, which is significantly different from the static planning methods in existing technologies; Enhance system-level robustness and usability: Continuously ensure high service satisfaction rates in extreme joint uncertainty scenarios, while avoiding resource waste in normal scenarios.

[0163] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for demand forecasting and layout optimization of electric vehicle charging stations, characterized in that, include: Step S1, Two-dimensional uncertainty quantification: Collect charging demand data and charging pile operation data of the target area, perform preprocessing and feature extraction to obtain demand feature set and equipment feature set; Based on the aforementioned demand feature set, the probability distribution and demand confidence interval of future charging demand fluctuations are predicted using a demand prediction model based on a Bayesian neural network. Based on the device feature set, the fault confidence interval of the charging pile's fault probability and fault duration in the future time period is predicted by the fault prediction model. Couple the demand confidence interval with the fault confidence interval to generate a set of joint uncertainty scenarios that includes multiple joint scenarios and their occurrence probabilities; Step S2, Solving for robust redundancy layout: Using the aforementioned set of joint uncertainty scenarios as constraints, a multi-objective optimization function is constructed with charging service satisfaction rate, resource utilization rate, and total life cycle cost as objectives. An improved non-dominated sorting genetic algorithm is used to solve the function. The algorithm dynamically allocates uncertainty weights based on the intensity of demand fluctuations and the degree of equipment aging. The intensity of demand fluctuations includes a scenario intensity index that characterizes the overall range of demand fluctuations. The scenario intensity index is calculated based on the joint uncertainty scenario set. Based on the solution results, the configuration scheme of redundant charging piles in the target area is determined by the redundancy configuration model, including the type, quantity and layout location of hot backup redundant piles and cold backup redundant piles.

2. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 1, characterized in that, The step S1, which involves collecting the charging demand data and charging pile operation data, specifically includes: Demand data collection: Real-time operating parameters of the regional power grid are obtained through load sensors; user behavior data, including travel trajectories and charging durations, are obtained through application programming interfaces (APIs) of vehicle terminals or traffic management platforms; and regional development dynamic data are obtained through planning data interfaces. Equipment data acquisition: The operating parameters of the core components are collected through test sensors deployed on the charging piles; historical fault records and related information are obtained through the charging pile operation and maintenance management system. Auxiliary data acquisition: Obtain the power grid load redundancy threshold through the power grid dispatching system interface, and obtain extreme weather early warning information through the meteorological data interface.

3. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 1, characterized in that, The step S1, which involves obtaining the demand confidence interval based on the demand forecasting model, specifically includes: Model training: The Bayesian neural network is trained using preprocessed historical charging demand data as supervision labels; Interval prediction: Based on the trained model, the feature data of the prospective period is used as input to output the predicted interval of charging demand at different confidence levels, and the uncertainty of demand fluctuation is represented in the form of a probability distribution. Scenario segmentation: Using the Monte Carlo simulation method, multiple demand scenarios covering different fluctuation intensities are sampled from the probability distribution to form an instantiation of the demand confidence interval.

4. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 1, characterized in that, The step S1 of generating the joint uncertainty scenario set specifically includes: Coupling operation: The demand fluctuation probability distribution represented by the demand confidence interval is coupled with the equipment failure probability distribution represented by the failure confidence interval to construct a joint probability model representing two-dimensional uncertainty; Scenario generation: Based on the aforementioned dual-dimensional joint probability distribution model, multiple joint scenarios with definite probabilities are generated through sampling calculation; wherein, each joint scenario is jointly defined by a combination of demand fluctuation range at a specific confidence level and equipment failure at a specific probability. Weighting and Constraint Extraction: Assign a weight to each generated joint scenario, which is equal to the probability of occurrence of the scenario in the two-dimensional joint probability distribution model; and identify extreme joint scenarios with a probability of occurrence below a preset threshold as hard constraints that must be satisfied in subsequent optimization solutions.

5. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 1, characterized in that, The improved non-dominated sorting genetic algorithm in step S2 dynamically allocates weights based on the intensity of demand fluctuations and the degree of equipment aging, specifically including: Feature quantification: Based on the joint uncertainty scenario set, calculate the scenario intensity index that characterizes the overall demand fluctuation range, and calculate the regional aging index that characterizes the overall health status of the equipment based on the fault confidence interval. Dynamic weight generation: In each iteration of the algorithm, based on the real-time values ​​of the scene intensity index and the area aging index, a combination of weight coefficients is dynamically generated through preset mapping rules to balance the relationship between service continuity objectives and cost reliability objectives. Weighting Application: The dynamically generated weight coefficients are combined and applied to the solution process of the multi-objective optimization function, so that the search direction of the algorithm can adaptively tend to alleviate the more prominent risk dimension in the current uncertainty features.

6. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 5, characterized in that, The construction and solution of the multi-objective optimization function in step S2 specifically includes: Function Construction: The constructed multi-objective optimization function is expressed as follows: ,in, To improve service satisfaction, For total lifecycle cost, For resource utilization, , , These are the preset target weights for service satisfaction rate, cost, and resource utilization rate; Dynamic coupling of coefficients: The values ​​of the service satisfaction rate target weight and the cost target weight are not fixed constants, but are coupled with the dynamically generated weight coefficient combination, so that the core trade-off relationship of the multi-objective optimization function can be adjusted in real time according to the system risk characteristics represented by the joint uncertainty scenario set generated in step S1. Solution set mapping output: The multi-objective optimization function is optimized by the improved non-dominated sorting genetic algorithm, and a scheme is selected from the final non-dominated solution set. The scheme directly outputs the redundant stub configuration parameters corresponding to the current weight tendency.

7. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 1, characterized in that, The determination of the number of redundant charging piles in step S2 is specifically achieved in the following way: Dynamic parameter acquisition: The key input parameters in the redundant configuration model—demand fluctuation rise rate and regional average failure probability—are taken in real time from specific statistical values ​​in the latest output of the demand confidence interval and the failure confidence interval in step S1. Redundancy requirement calculation: The dynamically acquired parameters are substituted into the redundancy configuration model. The logic of the redundancy configuration model is configured as follows: when the demand fluctuation rise rate increases significantly, the calculation result tends to increase the redundancy resources to prevent demand peaks; when the regional average failure probability increases significantly, the calculation result tends to increase the redundancy resources to resist equipment cluster failures. Strategy execution: The calculated total redundancy requirement, combined with the layout position obtained by the multi-objective optimization function, constitutes the configuration scheme.

8. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 1, characterized in that, The method further includes step S3, dynamic iterative optimization: After the implementation of the plan, dynamic demand and equipment operation data of the target area will be collected in real time. The demand forecasting model and the fault prediction model are updated dynamically based on the data to update the demand confidence interval, the fault confidence interval, and the joint uncertainty scenario set. The configuration scheme of the redundant charging piles is dynamically adjusted based on the updated scenario set and preset triggering conditions related to demand fluctuations or equipment failures. The real-time data acquisition in step S3 to drive the closed-loop iteration of the model and solution specifically includes: Feedback data collection: After the configuration scheme is implemented, the real-time operation feedback data of the target area is continuously acquired through the deployed data collection node network. The feedback data includes the actual charging demand sequence and the charging pile operation status sequence. Feedback feature processing: The real-time operation feedback data is processed in real time to extract dynamic demand features and dynamic equipment health features for model updates; Model Iteration Driven: The extracted dynamic demand features and dynamic equipment health features are used as incremental training data and input into the demand prediction model and fault prediction model in step S1 to drive the update of the demand confidence interval, the fault confidence interval and the joint uncertainty scenario set.

9. The method for demand forecasting and layout optimization of electric vehicle charging stations according to claim 8, characterized in that, The dynamic iterative optimization in step S3 is implemented through the following closed-loop process: Model and scenario set update: Based on the real-time operation feedback data, the demand prediction model and the fault prediction model are retrained in a rolling manner at a period not exceeding the preset update time, and the demand confidence interval, the fault confidence interval and the joint uncertainty scenario set are updated accordingly. Layout scheme re-optimization trigger: When the updated joint uncertainty scenario set indicates that the change in system risk characteristics reaches a set threshold, the robust redundancy layout solution process of step S2 is automatically triggered, with the updated scenario set as input. Dynamic deployment of the solution: The new configuration solution obtained by resolving the solution is compared with the current running solution, and an executable set of adjustment instructions is generated to complete the seamless switching and dynamic deployment from the old solution to the new solution.

Images